Plant Foods, Antioxidant Biomarkers, and the Risk of Cardiovascular Disease, Cancer, and Mortality: A Review of the Evidence

ABSTRACT

Although a high intake of plant foods such as fruits, vegetables, whole grains, nuts, and legumes has been recommended for chronic disease prevention, it has been unclear what is the optimal amount of intake of these foods and whether specific subtypes are particularly beneficial. The evidence from several recently published meta-analyses on plant foods and antioxidants and various health outcomes is reviewed as well as more recently published studies. In meta-analyses of prospective studies, inverse associations were observed between intake of fruits, vegetables, whole grains, and nuts and the risk of coronary artery disease, stroke, cardiovascular disease overall, total cancer, and all-cause mortality. The strongest reductions in risk were observed at an intake of 800 g/d for fruits and vegetables, 225 g/d for whole grains, and 15–20 g/d for nuts, respectively. Whole-grain and nut consumption was also inversely associated with mortality from respiratory disease, infections, and diabetes. Stronger and more linear inverse associations were observed between blood concentrations of antioxidants (vitamin C, carotenoids, vitamin E) and cardiovascular disease, cancer, and all-cause mortality than for dietary intake. Most studies that have since been published have been consistent with these results; however, further studies are needed on subtypes of plant foods and less common causes of death. These results strongly support dietary recommendations to increase intake of plant foods, and suggest optimal intakes for chronic disease prevention may be ∼800 g/d for intakes of fruits and vegetables, 225 g/d for whole grains, and 15–20 g/d for nuts. Diets high in plant foods could potentially prevent several million premature deaths each year if adopted globally.

Introduction

A high intake of plant foods including fruits, vegetables, whole grains, nuts, and legumes has long been recommended to the general population to reduce the risk of chronic diseases such as cardiovascular disease, cancer, and type 2 diabetes (1), which are among the main causes of premature death worldwide (2). For example, the 5 A Day for Better Health Program was launched in the United States in 1991 to increase the intake of fruits and vegetables to a minimum of 5 servings/d (3, 4). Similar campaigns have also been launched in many other countries (5, 6). Such recommendations have largely been based on the results from epidemiological studies which have consistently shown reductions in risk of coronary artery disease and stroke with a higher intake of fruits and vegetables (7, 8). In addition, it has been observed that whole grains have been associated with a reduced risk of coronary artery disease and type 2 diabetes (9, 10) and that a high intake of nuts has been associated with a reduced risk of coronary artery disease (11). Historically there was also strong evidence that a high intake of fruit and vegetables reduced the risk of several cancers and when the first report of the World Cancer Research Fund and the American Institute of Cancer Research, Food, Nutrition and the Prevention of Cancer: A Global Perspective (1), was published in 1997 it stated that there was convincing evidence that a high intake of fruit and/or vegetables reduced the risk of cancers of the mouth and pharynx, esophagus, lung, stomach, colon, and rectum, and that they probably reduced the risk of cancers of the larynx, breast, and bladder. However, the evidence for a benefit of fruit and vegetable intake in cancer prevention became weaker in the following decade as more prospective cohort studies accrued and showed weaker or no associations between fruit and vegetable intakes and the risk of several cancers (12–18). In fact, none of the associations were deemed convincing when an update of the previous report was published in 2007 (19). The main reason for the change in the conclusions was that the results of the 1997 report mainly came from retrospective case-control studies, which may have been affected by recall and selection biases to a greater degree than the later cohort studies. In addition, a more systematic and rigorous approach to the published data was used in the 2007 report than in the earlier report, incorporating systematic literature reviews and meta-analyses which quantified the association between each dietary factor and cancer risk. Nevertheless, the evidence was considered probable that fruit and/or vegetables protect against cancers of the mouth, pharynx, larynx, esophagus, stomach, and lung also in the Second Expert Report (19). Associations for which the evidence has been graded probable or convincing are strong enough for recommendations to be made. Additional cohort studies have emerged since the 2007 report (20–23) and these have been incorporated in the Continuous Update Reports which have since been published (24–29). This review provides a summary of the available data on intake of plant foods, antioxidants, and the risk of cardiovascular disease, cancer, type 2 diabetes, and all-cause and cause-specific mortality, as well as of the assessments of plant foods and cancer risk in the Third Expert Report which has just been published (30). The focus of the review is on recently published meta-analyses, but additional studies that have since been published have also been included.

Fruits and Vegetables

In a comprehensive systematic review and meta-analysis of 95 studies (142 publications) (31), we found summary RRs of 0.92 (95% CI: 0.90, 0.94, I² = 0%, n = 15) for coronary artery disease, 0.84 (95% CI: 0.76, 0.92, I² = 73%, n = 10) for stroke, 0.92 (95% CI: 0.90, 0.95, I² = 31%, n = 13) for cardiovascular disease, 0.97 (95% CI: 0.95, 0.99, I² = 49%, n = 12) for total cancer, and 0.90 (95% CI: 0.87, 0.93, I² = 83%, n = 15) for all-cause mortality per 200 g/d of fruit and vegetable intake (1 serving = 80 g) (Table 1). Similar associations were observed for fruits and vegetables when evaluated separately. In nonlinear dose–response analyses, the associations between total fruit and vegetable intake and coronary artery disease or mortality from stroke were linear up to 800 g/d, whereas for the remaining outcomes the associations were nonlinear. For stroke incidence and mortality combined, cardiovascular disease, and all-cause mortality the largest reductions were observed when increasing fruit and vegetable intake from 0 to 400 g/d, but some further reductions were observed up to 800 g/d, whereas for total cancer there was little further benefit beyond an intake of 600 g/d (31).

TABLE 1.

Summary RRs (95% CIs) from meta-analyses of intakes of plant foods and antioxidants and coronary artery disease, stroke, CVD, cancer, and all-cause mortality¹

	Fruits and vegetables (per 200 g/d)			Whole grains (per 90 g/d)			Nuts (per 28 g/d)			Vitamin C, diet (per 100 mg/d)			Vitamin C, blood (per 50 µmol/L)
	n	RR (95% CI)	I 2	n	RR (95% CI)	I 2	n	RR (95% CI)	I 2	n	RR (95% CI)	I 2	n	RR (95% CI)	I 2
Coronary artery disease	15	0.92 (0.90, 0.94)	0%	7	0.81 (0.75, 0.87)	9%	11	0.71 (0.63, 0.80)	47%	11	0.88 (0.79, 0.98)	65%	4	0.74 (0.65, 0.83)	0%
Stroke	10	0.84 (0.76, 0.92)	73%	6	0.88 (0.75, 1.03)	56%	11	0.93 (0.83, 1.05)	14%	12	0.92 (0.87, 0.98)	68%	4	0.70 (0.61, 0.81)	0%
CVD	13	0.92 (0.90, 0.95)	31%	10	0.78 (0.73, 0.85)	40%	12	0.79 (0.70, 0.88)	60%	10	0.89 (0.85, 0.94)	27%	6	0.76 (0.65, 0.87)	56%
Cancer	12	0.97 (0.95, 0.99)	49%	6	0.85 (0.80, 0.91)	37%	8	0.85 (0.76, 0.94)	42%	8	0.93 (0.87, 0.99)	46%	5	0.74 (0.66, 0.82)	0%
All-cause mortality	15	0.90 (0.87, 0.93)	83%	11	0.83 (0.77, 0.90)	83%	16	0.78 (0.72, 0.84)	66%	14	0.89 (0.85, 0.94)	80%	8	0.72 (0.66, 0.79)	48%
Respiratory disease	—	—	—	4	0.78 (0.70, 0.87)	0%	3	0.48 (0.26, 0.89)	61%	—	—	—	—	—	—
Diabetes	—	—	—	4	0.49 (0.23, 1.05)	85%	4	0.61 (0.43, 0.88)	50%	—	—	—	—	—	—
Infectious disease	—	—	—	3	0.74 (0.56, 0.96)	0%	2	0.25 (0.07, 0.85)	0%	—	—	—	—	—	—
Nervous system disease	—	—	—	2	1.15 (0.66, 2.02)	79%	3	0.65 (0.40, 1.08)	6%	—	—	—	—	—	—
Non-CVD/noncancer	—	—	—	5	0.78 (0.75, 0.82)	0%	—	—	—	—	—	—	—	—	—

¹CVD, cardiovascular disease.

When evaluating specific types of fruits and vegetables we found inverse associations between the intake of apples and pears, citrus fruits, green leafy vegetables and/or salads, and cruciferous vegetables, and risk of cardiovascular disease and all-cause mortality, whereas for total cancer inverse associations were observed for the intake of green-yellow vegetables and cruciferous vegetables (31). Under several assumptions, including that of a causal relation between fruit and vegetable intake and these outcomes, and based on the results from the nonlinear dose–response analysis, an estimated 5.6 and 7.8 million premature deaths may have been attributable globally in 2013 to a fruit and vegetable intake <500 g/d and <800 g/d, respectively (31).

Because of limited data at that time it was not possible to conduct analyses of fruit and vegetable intake and other causes of death because only the European Prospective Investigation into Cancer and Nutrition (EPIC) study had analyzed other causes of death (32). In the EPIC study, inverse associations were observed between fruit intake and mortality from digestive diseases (HR: 0.77; 95% CI: 0.66, 1.00) or unknown causes of death (HR: 0.88; 95% CI: 0.81, 0.97) and a positive association was observed for diseases of the nervous system (HR: 1.60; 95% CI: 1.22, 2.11), whereas for vegetables inverse associations were observed for mortality from circulatory diseases (HR: 0.78; 95% CI: 0.71, 0.87), respiratory diseases (HR: 0.78; 95% CI: 0.62, 0.97), digestive diseases (HR: 0.62; 95% CI: 0.47, 0.82), and for other causes of death (HR: 0.80; 95% CI: 0.66, 0.97) (32).

More recently, the China Kadoorie Biobank Study (462,342 participants and 17,894 deaths) published on fresh fruit intake and multiple causes of death and found inverse associations with most specific causes of death, including ischemic heart disease; stroke (total, ischemic, and hemorrhagic); other cardiovascular diseases; cancers of the esophagus, stomach, and colorectum; chronic obstructive pulmonary disease; respiratory disease; all other major chronic diseases; and all other causes of death, as well as all causes of death (33). No association was observed for lung or liver cancer or for transport accidents, and the latter could be considered as a negative control because one would not expect any association with deaths from transport accidents. Adjustments were made for age, sex, region, smoking, alcohol intake, education, income, consumption of meat, consumption of dairy products, consumption of preserved vegetables, survey season, physical activity, and BMI. Unfortunately, it was not possible to analyze vegetable intake because the highest frequency on the questionnaire was daily intake and 95% of the participants ate vegetables daily or more frequently.

The PURE Study, which is a global cohort study with 135,335 participants and 5796 deaths and with data from 18 low-, middle-, and high-income countries in North America, Europe, South America, the Middle East, South Asia, China, Southeast Asia, and Africa, also recently published on fruit, vegetable, and legume intake and risk of cardiovascular disease and mortality (34). In contrast to most other studies on fruit and vegetable intake and chronic disease and mortality risk (31), this study combined legume intake and fruit and vegetable intake in the primary analyses in the article. In this analysis there was no association between fruit, vegetable, and legume intake and risk of major cardiovascular events, myocardial infarction, or stroke, but an inverse association was observed for cardiovascular death, noncardiovascular death, and all-cause mortality, with the lowest risk observed at 5 to <6 servings/d for cardiovascular death (HR: 0.58; 95% CI: 0.42, 0.80), 3 to <4 servings/d for noncardiovascular death (HR: 0.77; 95% CI: 0.66, 0.89), and 3 to <4 servings/d for all-cause mortality (HR: 0.78; 0.69, 0.88) (1 serving was defined as 125 g/d, which is somewhat higher than the 80 g that had been used as a serving size previously). However, because legume intake was as strongly, if not more strongly, inversely associated with these outcomes as fruit and vegetable intake and had a much lower range of intake, the flattening of the dose–response curve at quite low intakes may at least partly be explained by an effect of legume intake. Interestingly, when fruits and vegetables were analyzed separately from legumes (reported in the online supplement), effect sizes and dose–response relations that were more consistent with the most recent meta-analysis (31) emerged, although some of the associations were still not statistically significant. Comparing an intake of 7 to <8 servings (the nadir of the dose–response curve for all-cause mortality) with <1 serving/d, the HRs were 0.81 (95% CI: 0.64, 1.03) for major cardiovascular disease, 0.87 (95% CI: 0.62, 1.22) for myocardial infarction, 0.80 (95% CI: 0.54, 1.17) for stroke, 0.60 (95% CI: 0.38, 0.95) for cardiovascular disease mortality, 0.73 (95% CI: 0.55, 0.96) for noncardiovascular disease mortality, and 0.69 (95% CI: 0.55, 0.86) for all-cause mortality (34).

A study from Taiwan (4176 participants, 1237 deaths) reported an inverse association between daily compared with nondaily intake of fruits and vegetables and cardiovascular and all-cause mortality (35). Other studies that have since been published include an Australian study which suggested inverse associations between intake of total vegetables, cruciferous vegetables, and allium vegetables and risk of coronary artery disease, stroke, and cardiovascular disease overall, although no significant association was observed for green leafy vegetables and yellow, orange, and red vegetables (36); a second study from Iran also reported an inverse association between allium vegetable intake and risk of cardiovascular disease (37). In the Osteoarthritis Initiative Cohort Study there was a positive association between fried potato intake and all-cause mortality (HR: 2.26; 95% CI: 1.15, 4.47) for ≥3/wk compared with ≤1 time/mo, but no association was observed for potatoes overall or unfried potatoes (38). In the 2017 meta-analysis there was an inverse association between high compared with low intake of potatoes and all-cause mortality (HR: 0.78; 95% CI: 0.74, 0.83, n = 4), but not in the dose–response analysis (HR: 0.91; 95% CI: 0.81, 1.03 per 100 g/d, n = 4), and the difference in the results may be at least partly due to differences in the studies included in each analysis (31). Because there was no association between intake of potatoes (per 100 g/d) and risk of coronary artery disease (HR: 0.99; 95% CI: 0.93, 1.05, n=6), stroke (HR: 0.98; 95% CI: 0.94, 1.02, n=4), cardiovascular disease overall (HR: 1.01; 95% CI: 0.97, 1.04, n=4), and total cancer (HR: 0.99; 95% CI: 0.95, 1.02, n=3), the inverse association observed in the high compared to low analysis of all-cause mortality could be due to chance or selective reporting, and is difficult to explain, except for if there would be an inverse association with other causes of death. However, such evidence is lacking. Although potatoes are not counted as part of the 5 recommended servings per day, further studies are needed to clarify the association between potato intake (and fried versus boiled or baked potatoes) and different health outcomes, particularly given the limited number of studies currently published.

Although the evidence regarding fruit and vegetable intake and cancer risk has become weaker over the last few decades, it is likely that there is an association at least with some cancers because there was a reduced risk of total cancer with a high intake of fruits and vegetables in the most recent meta-analysis (31). There was a 14% reduction in total cancer risk (summary RR: 0.86; 95% CI: 0.83, 0.89) when comparing people with an intake of 600 g fruits and vegetables/d with those eating only 40 g/d (31). With regard to subtypes, only cruciferous vegetables and green/yellow vegetables appeared to be protective; however, again there was a limited number of studies and we cannot exclude the possibility that other subtypes also may be protective (31). With regard to specific cancers, the WCRF Third Expert Report that came out in 2018 considered that there is probable evidence that fruits and vegetables reduce the risk of aerodigestive cancers as a group, but none of the individual cancer sites assessed had a judgment of probable or convincing any longer (30). For several individual cancers, the evidence is now considered limited and suggestive of an association, or limited and no conclusion is possible (24–30). Updated meta-analyses based on the Continuous Update Project have suggested inverse associations between fruit and vegetable intake and risk of cancers of the colorectum (39), breast (40), bladder (41), and lung (42); however, the associations were in general weak and only just significant. For lung and bladder cancer, residual confounding from smoking is difficult to exclude, particularly when associations are nonsignificant among never smokers (41, 42), and this may have been a major reason why the judgments were not stronger. However, more studies are needed with stratification for smoking status because statistical power is more limited among never smokers. The judgment of a probably causal relation between a dietary factor and cancer risk requires that one with confidence should be able to exclude the possibility that the observed association results from random or systematic error, including confounding, measurement error, and selection bias, and the criteria with regard to lack of confounding may therefore not have been fulfilled (30). Results from the Pooling Project of Prospective studies have suggested inverse associations between fruit and vegetable intake and risk of cancers of the lung (43) and kidney (44), but no association with colon (45), pancreatic (46), breast (except for an inverse association with estrogen receptor negative tumors) (47), ovarian (48), and prostate cancer (49). Some of the differences in the results between the Pooling Project of Prospective studies and the Continuous Update Project are likely due to differences in terms of which studies are included in the meta-analyses because not all the individual studies included in the Pooling Project have published separately (except for in the Pooling Project) on every cancer site, whereas the Continuous Update Project also includes other studies that may not have met the inclusion criteria of the Pooling Project. However, other differences including duration of follow-up, categorization of intakes, and inclusion of adjustment variables may also contribute to differences in findings. Although initial analyses of fruit and vegetable intake and breast cancer risk were largely null in both the EPIC study (18) (which was not included in the analysis of the Pooling Project) and the Nurses’ Health Study (16), an inverse association between fruit and vegetable consumption and breast cancer risk has become apparent with longer follow-up in both studies (50, 51).

Whether specific types of fruits and vegetables are particularly beneficial for individual cancers has also been explored in some meta-analyses and pooled analyses, with inverse associations observed between the intake of citrus fruits and bladder cancer risk (RR: 0.87; 95% CI: 0.76, 0.99) and between citrus fruits (RR: 0.91; 95% CI: 0.85, 0.98), cruciferous vegetables (RR: 0.92; 95% CI: 0.87, 0.98), and green leafy vegetables (RR: 0.89, 95% CI: 0.79, 1.00) and lung cancer risk (42). In addition, inverse associations were observed between the intake of bananas (RR: 0.88; 95% CI: 0.78, 0.99) and spinach (RR: 0.89; 95% CI: 0.82, 0.97) and colon cancer (45), and the intake of apples/pears (RR: 0.92; 95% CI: 0.85, 0.99), peaches/nectarines/apricots (RR: 0.81; 95% CI: 0.70, 0.94), strawberries (RR: 0.56; 95% CI: 0.41, 0.76), and lettuce (RR: 0.91; 95% CI: 0.84, 0.98) and estrogen receptor negative breast cancers, but no significant associations were observed for estrogen receptor positive tumors (47). Moreover, a slight inverse association was observed between lettuce and advanced (RR: 0.91; 95% CI: 0.85, 0.98) and fatal (RR: 0.86; 95% CI: 0.78, 0.94) prostate cancer, but a positive association was observed between corn and advanced (RR: 1.53; 95% CI: 1.12, 2.07) and fatal (RR: 1.49; 95% CI: 1.01, 2.20) prostate cancer (49). No significant associations were reported between subtypes of fruits and vegetables and ovarian cancer risk (48); however, slight positive associations were observed between the intake of strawberries (RR: 1.13; 95% CI: 1.01, 1.27), Brussels sprouts (RR: 1.26; 95% CI: 1.03, 1.54), green pepper (RR: 1.15; 95% CI: 1.01, 1.30), and tomatoes/tomato juice (RR: 1.05; 95% CI: 1.01, 1.09) and pancreatic cancer, but no subtypes were significantly inversely associated with risk (46). However, given the large number of comparisons, it is possible that some of these findings may have been due to chance and the lack of or limited mechanistic data that could explain these findings is another important limitation.

Whole Grains

In another meta-analysis (52), we found a reduced risk of coronary artery disease (summary RR: 0.81; 95% CI: 0.75, 0.87, I² = 9%, n = 7), cardiovascular disease (RR: 0.78; 95% CI: 0.73, 0.85, I² = 40%, n = 10), total cancer (RR: 0.85; 95% CI: 0.80, 0.91, I² = 37%, n = 6), all-cause mortality (RR: 0.83; 95% CI: 0.77, 0.90, I² = 83%, n = 11), and mortality from respiratory disease (RR: 0.78; 95% CI: 0.70, 0.87, I² = 0%, n = 4), infectious disease (RR: 0.74; 95% CI: 0.56, 0.96, I² = 0%, n = 3), and noncardiovascular, noncancer causes of death (RR: 0.78; 95% CI: 0.75, 0.82, I² = 0%, n = 5) per 90 g or 3 servings/d (1 serving = 30 g) of whole-grain intake (Table 1). Some suggestion of a reduced risk was also observed for stroke (RR: 0.88; 95% CI: 0.75, 1.03, I² = 56%, n = 6) and mortality from diabetes (RR: 0.49; 95% CI: 0.23, 1.05, I² = 85%, n = 4), but the associations were only significant in the nonlinear dose–response analysis (52). Nonlinear associations were observed in all of the analyses with the exception of total cancer, and slightly stronger associations were observed when increasing whole-grain intake from 0 to between 50 and 100 g/d, than at higher intakes (52). However, for coronary artery disease, total cancer, all-cause mortality, and mortality from respiratory diseases and noncardiovascular, noncancer causes of death, there were further reductions in risk up to an intake of 225 g/d, which is equal to 7.5 servings or 7.5 slices of whole-grain bread per day (52). This was the highest amount of intake across studies so we were not able to draw any conclusions with regard to the health effects of even higher intakes. This is similar to the amount of whole-grain intake recommended by the Scandinavian countries (70–90 g dry weight ∼200–250 g of whole-grain products) (53), but considerably higher than that recommended in the United States (54). For specific types of whole-grain products, we found inverse associations between intake of whole-grain bread, whole-grain breakfast cereals, and bran and the risk of coronary artery disease and cardiovascular disease, between intake of whole-grain bread and total cancer, and between intake of whole-grain bread and whole-grain breakfast cereals and all-cause mortality. No association was observed between intake of refined grains or rice (total, white, or brown) and risk of coronary artery disease, stroke, or cardiovascular disease, but some evidence of slight inverse associations was observed for the intake of total grains and refined grains in relation to total cancer and all-cause mortality. However, all the latter results were based on a limited number of studies and further data from additional studies are needed. In any case, the inverse associations for total grains and total cancer and all-cause mortality appear to be largely driven by intake of whole grains because the few inverse associations observed for intake of refined grains were much weaker than those observed for whole grains (52) and, in addition, the associations between total grain intake and coronary artery disease, stroke, and cardiovascular disease were null.

We also previously reported inverse associations between the intake of whole grains and the risk of type 2 diabetes incidence with a 32% reduction in risk per 90 g/d, but no association was observed for refined grains (10). There was little evidence of further benefit with intakes >90 g/d in the nonlinear dose–response analysis (10). Total grains, whole-grain bread, whole-grain breakfast cereals, wheat bran, and brown rice were all inversely associated with type 2 diabetes incidence, but no association was observed for wheat germ, while a positive association was observed for white rice in the dose-response analysis (10). The findings are further supported by evidence from a meta-analysis on fiber intake and risk of type 2 diabetes, which found stronger inverse associations between cereal fiber intake and type 2 diabetes than with other specific fiber sources (55). As part of the Continuous Update Project we also reported an inverse association between intake of whole grains and colorectal cancer risk with a summary RR of 0.83 (95% CI: 0.78, 0.89, I² = 18%, n = 6) per 90 g/d (56) and based on these analyses the evidence that whole grains reduce colorectal cancer risk was recently updated to probable in the Continuous Update Report from 2017 (30) and in the Third Expert Report (57). There are only limited cohort data regarding whole-grain intake and risk of other cancers. A few studies have suggested inverse associations with cancers of the upper aerodigestive tract (58, 59), small intestine (60), liver (61), and kidney (62), but studies on hormonal cancers including breast (63, 64), prostate (65, 66), and endometrial cancer (67) are largely null.

Nuts

In a meta-analysis of nut intake and various health outcomes (68), the summary RRs per 28 g/d (1 serving = 28 g) increase in nut intake were as follows: for coronary artery disease, 0.71 (95% CI: 0.63, 0.80, I² = 47%, n = 11); stroke, 0.93 (95% CI: 0.83, 1.05, I² = 14%, n = 11); cardiovascular disease, 0.79 (95% CI: 0.70, 0.88, I² = 60%, n = 12); total cancer, 0.85 (95% CI: 0.76, 0.94, I² = 42%, n = 8); all-cause mortality, 0.78 (95% CI: 0.72, 0.84, I² = 66%, n = 15); and for mortality from respiratory disease, 0.48 (95% CI: 0.26, 0.89, I² = 61%, n = 3); diabetes, 0.61 (95% CI: 0.43, 0.88, I² = 0%, n = 4); neurodegenerative disease, 0.65 (95% CI: 0.40, 1.08, I² = 5.9%, n = 3); infectious disease, 0.25 (95% CI: 0.07, 0.85, I² = 54%, n = 2); and kidney disease, 0.27 (95% CI: 0.04, 1.91, I² = 61%, n = 2) (Table 1). Similar results were found for peanuts and tree nuts. The associations between nut intake and these health outcomes were nonlinear and in most of the analyses there was no further benefit with an intake beyond 15–20 g/d. Under the assumption of a causal relation between nut consumption and reduced mortality we estimated that approximately 4.4 million premature deaths might have been attributable to a nut intake <20 g/d in 2013 globally (with the exception of Africa and the Middle East, areas for which we did not have data on nut intake) (68). In the Swedish Mammography Cohort and the Cohort of Swedish Men, it was recently reported that high nut intake was associated with a reduced risk of nonfatal myocardial infarction, atrial fibrillation, and possibly abdominal aortic aneurysm, although no association was observed for fatal myocardial infarction, heart failure, aortic valve stenosis, ischemic stroke, or intracerebral hemorrhage (69). The inverse association with nonfatal myocardial infarction and the lack of association with the risk of stroke are consistent with the findings from our meta-analysis; however, one limitation of this study was that the highest intake category for nuts was only ≥3 times/wk, whereas many previous studies had a range of intake of up to 1 serving/d or higher (68). Although randomized trials also have provided support for a benefit of nut consumption with regard to reductions in blood concentrations of total cholesterol, LDL cholesterol, apoB, and TGs (70), the dose–response analysis suggested that there was little or no reduction in total and LDL cholesterol with nut intakes ≤20–30 g/d and the lipid-lowering effects were more apparent at intakes of 60–100 g/d, which is slightly in contrast to the findings of our meta-analysis which found no or little further benefit in reducing risk of chronic diseases and mortality with an intake beyond 15–20 g/d. However, the top range of the intake across studies was 28 g/d (1 serving/d) in our meta-analysis and with the current epidemiological data it is not possible to say whether intakes beyond 1 serving/d can provide further reductions in risk. Given the limited number of very high nut consumers in most populations, very large studies would probably be needed to clarify this question.

Data regarding nut intake and risk of incident type 2 diabetes have largely shown null results (71–74); however, the possibility that specific types of nuts such as walnuts may be beneficial (75) needs further exploration. In addition, because there is some evidence suggesting that nuts may reduce weight gain over time (76), additional studies should provide results both with and without adjustment for BMI to clarify whether part of an association is mediated by reduced adiposity as suggested by one study (75). Nevertheless, some studies have suggested that nut consumption is associated with a reduction in risk of cardiovascular disease and all-cause mortality in patients with type 2 diabetes (77, 78) and randomized trials have also suggested benefits of nut consumption on cardiovascular disease risk factors (79–82), thus it may be beneficial for diabetes patients to increase their intake of nuts to prevent some of the complications of diabetes.

Very few cohort studies have been published on the association between nut intake and risk of specific cancers to date. A cohort study found a 32% reduction in risk of pancreatic cancer among women eating nuts ≥2 times/wk compared with those eating nuts never or almost never (83) and another cohort found a nonsignificant association in the same direction (84), whereas a smaller study found no association (85). A few prospective studies have suggested an inverse association between nut consumption and colorectal cancer (86–88); however, only two of these found significant associations (86, 88). A few cohort studies reported inverse associations between nut intake and stomach cancer (89–91), with stronger associations for gastric noncardia cancer than for gastric cardia cancer (90, 91). For breast (92–96) and prostate cancer (97) the available data show no clear association. One large cohort study and a case-control study also recently reported inverse associations between nut intake and lung cancer, with 14% and 26% reductions in the RR, respectively (98).

Legumes

A meta-analysis of prospective studies (99) found a reduced risk of coronary artery disease with a higher legume intake (RR: 0.86; 95% CI: 0.78, 0.94, per 4 servings/wk), but no association was observed for stroke (RR: 0.98; 95% CI: 0.84, 1.14) and similar results were found in a second meta-analysis (100). Data regarding legume intake and risk of type 2 diabetes have been largely null (101–105); however, the possibility that specific subtypes of legumes may be beneficial cannot be excluded (106, 107). Similarly, studies on legume intake and all-cause mortality have largely been null (108–111); however, a few studies suggested significant inverse associations (34, 112). Some studies have suggested inverse associations between intake of soy products and the risk of breast cancer (113, 114) and prostate cancer (115, 116); however, challenges have been observed in summarizing the available evidence owing to differences in the reporting between studies, because studies have reported on a range of soy-related exposures including soy protein, soy isoflavones, specific soy foods (e.g., tofu, soy milk), or total intake of soy foods (114). One meta-analysis suggested that the inverse association between soy intake and breast cancer was restricted to Asian populations where soy intake is much higher than in European and American populations (113); however, the limited number of cohort studies is a limitation. More research is needed before firm conclusions can be made and any additional studies could contribute to more definitive answers by providing more detailed results for intake of soy foods overall as well as for specific soy foods and constituents.

Antioxidant Biomarkers

Several components of plant foods, including fiber, minerals, and antioxidants such as flavonoids, vitamin C, carotenoids, and vitamin E, have been hypothesized to contribute to the beneficial associations observed between the intake of plant foods and a range of health outcomes. Several studies have assessed the association between the dietary intake or blood concentrations of vitamin C, carotenoids, and vitamin E and risk of cardiovascular disease, cancer, and premature mortality (117–119). Blood concentrations of vitamin C and carotenoids are considered to be biomarkers of fruit and vegetable intake (120, 121) and analyses using biomarkers might further advance our understanding of the relation between diet and chronic disease risk. In a meta-analysis based on data from the Continuous Update Project there was a very weak association between dietary carotenoid intake and the risk of breast cancer; however, studies that measured blood concentrations of carotenoids showed clear inverse associations, much stronger than those in the studies that assessed dietary intake using questionnaires (122). For example, there were 2–5% reductions in risk with increased dietary intake of carotenoids, β-carotene, and α-carotene; however, in the biomarker-based analysis there was an 18–26% reduction in risk with increased blood concentrations of these antioxidants (122).

In a further meta-analysis we assessed the associations of dietary vitamin C, carotenoids, and vitamin E, as well as blood concentrations of these antioxidants, with the risk of cardiovascular disease, total cancer, and all-cause mortality (123). Inverse associations were observed between dietary intake of vitamin C and carotenoids and most of these outcomes (although some variation in results existed between exposures and outcomes); however, when analyses used the blood-based biomarkers of the same antioxidants, stronger and more linear dose–response relations were often observed (123). Most of the studies on dietary intake of antioxidants reported on dietary intake from foods, but a few reported on intake from foods and supplements combined. Another meta-analysis suggested higher intake of dietary flavonoids and certain subtypes of flavonoids (flavones, flavanones, anthocyanidins) is associated with reduced risk of cardiovascular disease and all-cause mortality (124). In contrast, a large number of randomized trials have shown that use of antioxidant supplements (β-carotene, vitamin A, vitamin C, vitamin E, and selenium) has no benefits in the prevention of cardiovascular disease, cancer, or mortality, and in some cases may even increase risk (β-carotene, vitamin A, and vitamin E and all-cause mortality) (125–127). Although there is a possibility that antioxidant supplements may have some benefit in undernourished populations (128), this benefit may diminish over time (129), and the overall evidence suggests no benefit in well-nourished European and North American populations (125–127). At the same time, there is relatively consistent evidence across geographic locations (Europe, North America, and Asia) that high intakes of fruits and vegetables, and high dietary intakes and blood concentrations of fruit and vegetable–related nutrients, such as vitamin C and carotenoids, are associated with lower risk of cardiovascular disease, cancer, and mortality (31, 123). Altogether it therefore seems less likely that vitamin C or carotenoids are the components responsible for the reduction in the risk of cardiovascular disease, cancer, and mortality observed in these meta-analyses, but more likely these components are biomarkers of fruit and vegetables, which contain a myriad of beneficial components that may act synergistically to reduce the risk of these outcomes.

Mechanisms

Plant foods contain many nutrients and components that may contribute to a lower risk of chronic diseases and premature mortality including fiber, vitamin C, carotenoids, antioxidants, potassium, magnesium, flavonoids, unsaturated fats, vegetable protein, and possibly other compounds (130–132). Although some of these components may be particularly important in reducing chronic disease risk, it is also likely that the many compounds of plant foods act synergistically through several different mechanisms to reduce the risk of chronic diseases and mortality (133–135). A high intake of dietary fiber, fruits and vegetables, nuts, legumes, and whole grains has been found to reduce cholesterol concentrations, blood pressure, and inflammation, to improve vascular function, and to regulate the immune system (130, 136–140).

A meta-analysis of two randomized trials found a 3-mm Hg (95% CI: 1.09, 4.92 mm Hg) lower systolic blood pressure among participants who received dietary advice to eat more fruits and vegetables, but the association with diastolic blood pressure was nonsignificant and associations with blood lipids (total, LDL, and HDL cholesterol; TGs) were weak and nonsignificant (141). In general, a 1-mmol/L (=38.67 mg/dL) reduction in total cholesterol and LDL cholesterol reduces risk of ischemic heart disease by ∼30% (142), whereas a 20-mm Hg reduction in systolic blood pressure is associated with a 45% reduction in risk of ischemic heart disease and ∼50% reduction in risk of stroke (143). A meta-analysis of 12 intervention studies on tomato intake and cardiovascular disease risk factors found a 4.63-mg/dL (95% CI: 0.02, 9.24 mg/dL) reduction in LDL cholesterol and a 5.60-mm Hg decrease in systolic blood pressure among intervention studies using tomatoes or lycopene supplements as a treatment (144). Another meta-analysis of 45 randomized controlled trials on berries and cardiovascular disease risk factors (145) found LDL cholesterol significantly reduced by 0.14 mmol/L (95% CI: 0.03, 0.25 mmol/L), HDL cholesterol increased by 0.048 mmol/L (95% CI: 0.02, 0.08 mmol/L), TGs reduced by 0.07 mmol/L (95% CI: 0.003, 0.14 mmol/L), systolic blood pressure reduced by 2.07 mm Hg (95% CI: 0.64, 3.50 mm Hg), and diastolic blood pressure reduced by 1.43 mm Hg (95% CI: 0.39, 2.48 mm Hg). These results are partly consistent with the inverse associations observed between intake of tomatoes and risk of coronary artery disease and between intake of berries and risk of all-cause mortality (31), although not with the null association which was observed between intake of berries and cardiovascular disease; however, because of the limited number of studies more data are needed on fruit and vegetable subtypes. Randomized trials have also found suggestive evidence that intake of apples reduces total and LDL cholesterol (146, 147), VLDL cholesterol, and TGs (148) and may improve endothelial function (149). A trial of people with high normal blood pressure or hypertension who were randomly assigned to eat 3 kiwifruit/d compared with a control group eating 1 apple/d found a reduction in systolic blood pressure of 3.6 mm Hg (95% CI: 0.7, 6.5 mm Hg) and in diastolic blood pressure of 1.9 mm Hg (95% CI: 0.3, 3.6 mm Hg) among those who received the kiwifruit intervention. A randomized double-blind crossover trial using a freeze-dried grape polyphenol powder in people with metabolic syndrome found a 6-mm Hg reduction in systolic blood pressure and increased flow-mediated dilatation of 1.7 mm, but no difference in cholesterol, plasma glucose, or adiposity measures (150); however, a second trial found no benefit of grape juice on ambulatory blood pressure, but some reduction in nocturnal dip in systolic blood pressure and blood glucose (151). Some evidence suggests that more extreme amounts of fruit and vegetable intakes may lead to stronger reductions in cardiovascular disease risk factors (152, 153). The beneficial effect of some fruits and vegetables on blood cholesterol may be due to their high content of fiber which can bind bile salts in the small intestine and lead to fecal excretion of cholesterol, reduced glycemic response resulting in lower insulin stimulation of hepatic cholesterol synthesis, and fermentation of dietary fiber to SCFAs which may suppress cholesterol synthesis in the liver (154). The reduced blood pressure observed with higher intake of fruits and vegetables may be due to the high potassium content which increases urinary excretion of sodium, vasodilatation, and the glomerular filtration rate and decreases renin, renal natrium reabsorption, reactive oxygen species production, and platelet aggregation (155). However, other components including anthocyanins and flavonoids may also reduce blood pressure by increasing endothelium-dependent microvascular reactivity and plasma nitric oxide, and reducing C-reactive protein and E-selectin (136, 156, 157). In summary, there is a growing body of evidence showing that high intake of fruits and vegetables and specific types of fruits and vegetables reduces cardiovascular disease risk factors such as total cholesterol, LDL cholesterol, and systolic blood pressure and may improve endothelial function. Further studies are needed to clarify whether specific subtypes are particularly beneficial in reducing cardiovascular disease risk factors, and attention needs to be paid to what is chosen as the control diet.

A meta-analysis of 38 randomized controlled trials showed a 3.6-mg/dL (95% CI: 2.9, 4.4 mg/dL) reduction in total cholesterol, 4.2-mg/dL (95% CI: 3.4, 5.0 mg/dL) reduction in LDL cholesterol, and 4.2-mg/dL (95% CI: 2.6, 5.7 mg/dL) reduction in apoB per 1 serving (28.4 g) tree nuts/d (70); however, in the dose–response analysis there were stronger reductions at higher intakes and an intake of 100 g/d was associated with a reduction of total and LDL cholesterol of 25 mg/dL and 15–20 mg/dL, respectively. A previous pooled analysis of 25 randomized trials showed that an intake of nuts of 67 g/d was related to a 10.9-mg/dL reduction in total cholesterol, a 10.2-mg/dL reduction in LDL cholesterol, and improved ratios of LDL and total cholesterol to HDL cholesterol (137). The beneficial effects of nut consumption on blood cholesterol may to a large degree be driven by the content of unsaturated fatty acids; however, it has been shown that the cholesterol-lowering effect of nuts is 25% greater than what can be predicted based on their fatty acid content (134), thus it seems that other components of nuts also may be of importance.

A meta-analysis of randomized controlled trials on whole-grain intake and blood lipids found a significant 0.09 mmol/L (95% CI: 0.03, 0.15 mmol/L) reduction in LDL cholesterol and a 0.12 mmol/L (95% CI: 0.05, 0.19 mmol/L) reduction in total cholesterol, but no effect on HDL cholesterol or TGs (158). However, the reductions in lipids were stronger for oats, with reductions of 0.17 mmol/L (95% CI: 0.10, 0.25 mmol/L) for LDL cholesterol, 0.22 mmol/L (95% CI: 0.11, 0.32 mmol/L) for total cholesterol, and 0.14 mmol/L (95% CI: 0.05, 0.22 mmol/L) for TGs, whereas no effect was observed for wheat and mixed grains. One limitation of the results was that there was no dose–response relation between increasing whole-grain consumption and blood lipids; however, this may have been confounded by type of grain because the studies with higher intakes were predominantly studies using wheat or mixed grains. Few studies had assessed rye, barley, and rice so more studies are needed on those items before conclusions can be made; however, the lack of association between rice consumption and most lipids is consistent with the null association between intake of rice and risk of cardiovascular disease (52, 159). One of the few epidemiological studies that have investigated different sources of whole grains found as strong inverse associations for whole-grain wheat as for oat and rye intake in relation to risk of coronary artery disease, mortality, and type 2 diabetes (160, 161), but stronger inverse associations between whole-grain wheat and colorectal cancer than for oats and rye (162). Oats contain more soluble fiber (particularly β-glucans) than wheat, and this may explain the greater lipid-lowering effect of oats than of wheat, whereas wheat contains more insoluble fiber than oats, which adds bulk to the stool and helps the stool pass more quickly through the intestines.

Oxidative stress has been implicated in several chronic diseases including cardiovascular disease, cancer, diabetes, neurodegenerative disease, lung disease, and kidney disease (163). Reactive oxygen and nitrogen species are formed endogenously as a result of normal cellular and metabolic reactions, and oxidative stress refers to an imbalance between the production of reactive oxygen and nitrogen species and the antioxidant defense leading to oxidative damage that can threaten the normal function of the cell or organism (164). In the human body, antioxidants may act in a stepwise fashion or in antioxidant networks where several antioxidants are needed to convert free radicals to less active radicals (164). Increasing evidence suggests that antioxidants and other phytochemicals may act synergistically (165) and, if further confirmed, this could explain why supplements with one or a few antioxidants have not been shown to have any benefit in relation to chronic disease prevention and also why whole foods with many different antioxidants and other phytochemicals in more natural doses have substantial health benefits. Intervention studies using high intakes of fruits and vegetables alone (133) or combined with other lifestyle changes (166) have shown that intake of fruit and vegetables affects gene expression toward increased cellular stress defense (133) and modulation of tumorigenesis, protein metabolism and modification, intracellular protein traffic, and protein phosphorylation (166). In screening studies of the total antioxidant content of different foods it was found that berries, walnuts, pecans, sunflower seeds, and spices were among the foods with the highest content of antioxidants (167, 168). It is unclear to what degree the antioxidant content of different plant foods contributes to their health benefits or whether other constituents like fiber, unsaturated fatty acids, or other phytochemicals are equally important. Interestingly, some of the specific plant foods that were found to be beneficial in reducing risk of cardiovascular disease, cancer, and mortality in our meta-analyses, like whole grains, peanuts, apples, citrus fruits, cruciferous vegetables, green leafy vegetables, and tomatoes (31, 52, 68), are not among the plant foods with the highest antioxidant content (167, 168). Intake of berries, which were among the foods with the highest antioxidant content, was not more strongly associated with reduced mortality than intake of citrus fruits or apples (31). In contrast, in the PREDIMED study walnuts were associated with a stronger reduction in risk of cancer death than other nuts, whereas differences for total and cardiovascular disease mortality were smaller (169), and in the Nurses’ Health Study 1 and 2, but not in the Health Professionals Follow-up Study, walnuts appeared to be slightly more strongly associated with reduced cardiovascular disease risk than other nuts (170). However, it is unclear if this is due to the antioxidant content of walnuts, other constituents, or their combined effect. Altogether, this might suggest that other constituents of plant foods perhaps may be as important as antioxidants; however, given the limited number of epidemiological studies available with detailed data on plant foods with a high antioxidant content, much more work is needed to clarify these questions. Another important question is whether there are specific combinations of plant foods that may be particularly beneficial.

The finding that cruciferous vegetables was one of the few specific vegetable items that was associated with reduced risk of cancer overall (31), and in addition was associated with reduced risk of lung (42), bladder (41), and kidney cancer (broccoli) (44), is intriguing, because cruciferous vegetables are high in sulforaphane, a compound that can inhibit phase 1 enzymes which are responsible for activation of procarcinogens, and induce phase 2 enzymes which are critical in mutagen elimination (171). Whole grains are important sources of dietary fiber, which is fermented by the intestinal bacteria to SCFAs including acetate, butyrate, and propionate. Butyrate has been shown to downregulate tumor-related signaling pathways including the MAPK pathway, Wnt pathway, insulin pathway, and the vascular endothelial growth factor pathway (172) and to protect against experimental colorectal cancer through inhibition of histone deacetylase and reduced apoptosis and cell proliferation (173, 174). The SCFAs also reduce intestinal pH, which inhibits the conversion of primary bile acids to secondary bile acids and reduces the solubility of free bile acids and their carcinogenic potential (175). Fiber also increases fecal bulk, reduces the transit time, and therefore reduces the time potential for carcinogens to interact with the colonic epithelial cells (175). Nuts contain several constituents including ellagic acid (walnuts), anacardic acid (cashews), genistein (hazelnuts, peanuts), resveratrol (peanuts), inositol (cashews, peanuts), and fiber (all nuts) that could reduce cancer risk by inducing cell cycle arrest and apoptosis, or inhibiting cell proliferation, migration, invasion, angiogenesis, and metastasis (176–180).

There is some evidence that plant foods may reduce the risk of developing overweight and/or obesity and may reduce weight gain over time (76, 181–183). Although plant foods may prevent excess weight gain (76) and excess weight is an important risk factor for a large number of diseases and mortality (30, 184–190), it seems many of the observed associations between plant foods and chronic diseases and mortality persist even after adjustment for adiposity (31, 52, 68). Studies have also suggested a protective effect of plant foods, in particular whole grains, on the risk of type 2 diabetes (10), and type 2 diabetes is an established risk factor for a number of chronic diseases and causes of death (191). In the China Kadoorie Biobank Study the association between fresh fruit intake and cardiovascular death (HR: 0.63; 95% CI: 0.56, 0.72) was slightly, but not substantially, attenuated by stepwise additional adjustment for baseline BMI and waist circumference (HR: 0.64; 95% CI: 0.56, 0.72), systolic blood pressure (HR: 0.68; 95% CI: 0.60, 0.77), and blood glucose (HR: 0.70; 95% CI: 0.61, 0.79) (192), suggesting that only a small or modest part of the association may be explained by these risk factors. However, further cohort studies with repeated measures of both diet and cardiovascular disease risk factors are needed to formally test whether there is a temporal relation between intake of plant foods and changes in cardiovascular disease risk factors and whether such changes may mediate part of the benefit of plant foods on risk of chronic diseases and mortality.

Several studies have suggested that plant foods also may modulate the microbiota (193–198), although not all were consistent (199), and an increasing number of studies are linking the microbiota with a growing number of diseases (200, 201). Interestingly, a recent study in mice found that when diets were deprived of fiber, the gut bacteria started to break down the mucus layer of the intestines as a source of nutrients, leading to increased permeability through the intestines of pathogenic bacteria predisposing the mice to infections (202). This mechanism might explain the inverse association observed between the intake of whole grains, nuts, and fiber and risk of infectious disease mortality (52, 68, 203). Nevertheless, further studies are needed to clarify the underlying mechanisms observed for less common causes of death (31, 68).

Limitations of the Current Data

Confounding by other dietary factors and other lifestyle factors is a major issue and is difficult to completely rule out because people who eat more plant foods also tend to smoke less, be more physically active, have a lower prevalence of overweight or obesity, and eat less red and processed meat and fast foods. Many of the included studies adjusted for the most important confounding factors such as tobacco smoking, alcohol consumption, overweight and obesity, and physical activity and some also adjusted for other dietary factors (31, 52, 68). In general, there were few substantial differences between subgroups that adjusted for these confounding factors and those that did not (31, 52, 68). However, the possibility of residual confounding or the possibility that there are more specific confounders that may not have been adequately adjusted for in relation to specific causes of death cannot entirely be excluded. Nevertheless, in the Nurses’ Health Study and the Health Professionals Follow-up Study, the inverse associations between whole-grain intake and nut intake and mortality outcomes persisted in a number of subgroup analyses stratified by smoking, alcohol, physical activity, BMI, and red and processed meat intake (204, 205). In the EPIC study the inverse association between fruit and vegetable intake and all-cause mortality was not observed in never smokers (111), which could indicate confounding by smoking. However, in 2 other cohort studies an inverse association was observed between fruit and vegetable intake and all-cause mortality also in nonsmokers or never smokers (206, 207) and 2 studies (3 publications) (108, 208, 209) among Seventh Day Adventists, which are mainly nonsmoking and nondrinking populations, also reported inverse associations between fruit and/or vegetable intake and all-cause mortality. In the China Kadoorie Biobank Study, inverse associations were observed between fresh fruit intake and mortality from cardiovascular disease, cancer, and chronic obstructive pulmonary disease across most strata of age, sex, education, income, alcohol intake, smoking status, physical activity, preserved vegetable intake, BMI, and systolic blood pressure and there were few significant interactions in these stratified analyses (33). Thus, it seems the weight of the current evidence suggests that confounding by these risk factors does not fully explain the relation between plant food intake and overall health, although there may be specific conditions where confounding is more of an issue than others (e.g., smoking and lung cancer). Any further studies on plant foods and morbidity and mortality should report more detailed results with analyses stratified by other risk factors to better be able to exclude bias due to confounding.

Because intake of plant foods seems to be associated with a reduced risk of a number of outcomes, it could be argued that there is a lack of specificity in the results. However, first of all by taking a closer look at the relations observed, there appears to be some level of specificity. For example, with regard to specific cancer sites it seems that the intake of fruits and vegetables, whole grains, and nuts is more strongly associated with reductions in the risk of cancers of the digestive tract rather than with hormonally related or other nondigestive cancers (33). This gives the results some level of plausibility because plant foods are in direct contact with the digestive tract and therefore are in physical proximity to the organs most affected. In addition, the observation of no significant relation between fresh fruit intake and the risk of traffic accidents (33), an outcome where no relation would be expected, could be considered as a negative control. Second, specificity is a less important criterion for causality than many other criteria. This is because there are many other examples of lifestyle and metabolic risk factors including smoking (210, 211), obesity (184–190, 212–216), physical activity (186, 217–222), and diabetes (191) that are established as risk factors for a very wide range of diseases and causes of death. Given that low dietary intakes of plant foods, as well as other food groups, are important predictors of adiposity and type 2 diabetes, it could be argued that relations with complications of obesity and type 2 diabetes also are likely (223), but this is not to say that these are the only or even the most important mediators of the association between intake of plant foods and various health outcomes. Given that these associations persisted and were even equally strong among studies with adjustment for BMI or diabetes in our meta-analyses, it is likely that other mechanistic pathways may be more important than reduced adiposity and reduced insulin resistance in explaining these findings. However, because these studies only considered BMI and diabetes at baseline (cross-sectionally) and not during follow-up, it is not possible to make any conclusions in this regard because mediation analyses require a temporal relation between the exposure, the mediator, and the outcome.

Changes in Diet during Follow-Up and Measurement Errors

Most epidemiological studies to date have used a simple baseline dietary assessment under the assumption that dietary habits track well over time and reflect long-term dietary intake. However, it has been shown in a number of studies that dietary intake of specific food groups can change considerably over time (224). Not only do dietary habits change in healthy individuals, but patients with particular diagnoses such as type 2 diabetes or cardiovascular disease may change their diet in an attempt to treat or control their condition. This can be of particular importance when examining mortality outcomes because incident disease or metabolic risk factors (hypertension, elevated serum cholesterol) can trigger dietary changes that may influence the survival after diagnosis and therefore affect the diet–mortality relation. If only a baseline registration is utilized, dietary changes during follow-up will not be picked up and this can lead to regression dilution bias or bias toward the null, possibly attenuating the association between a dietary factor and all-cause or cause-specific mortality. The Nurses’ Health Study 1 and 2 and the Health Professionals Follow-up Study are some of the few studies that have collected updated dietary assessments during follow-up (225). In an analysis of the Nurses’ Health Study it was shown that individuals who were diagnosed with diabetes and hypercholesterolemia, but not hypertension, or who had undergone coronary artery bypass grafting or percutaneous coronary intervention, or who were diagnosed with angina during follow-up, increased their intake of cereal fiber after the diagnosis (226). When using only the baseline assessment to analyze the association between cereal fiber intake and risk of coronary artery disease, the HR for the highest quintile was 0.75 (95% CI: 0.65, 0.86), whereas in an analysis using cumulative averages of intake and no longer updating the questionnaires when the participants reported intermediate outcomes (coronary artery bypass grafting, percutaneous coronary intervention, or angina, diabetes, hypertension, or hypercholesterolemia), because patients may have altered their cereal fiber intake after diagnosis, the HR was 0.61 (95% CI: 0.52, 0.71) (226). Similarly it was shown that a diagnosis of coronary artery disease, stroke, and diabetes led to subsequent changes in whole-grain intake (204) and the HR for the association between whole-grain intake and cardiovascular disease mortality when using only the baseline dietary assessment was 0.93 (95% CI: 0.86, 1.00); however, using cumulative updated averages of repeated dietary assessments, the HR for the highest quintile was 0.85 (95% CI: 0.78, 0.92) (204). Considerable differences in the HRs have also been reported from these cohort studies when evaluating the association between red meat intake and type 2 diabetes and mortality by comparison of analyses using cumulative updated averages of repeated dietary assessments with analyses using only baseline dietary assessments (224, 225). In the China Kadoorie Biobank Study, HRs were strengthened after corrections were made for regression dilution bias. The observed HRs per 1 daily portion of fresh fruit were 0.78 (95% CI: 0.73, 0.84) for cardiovascular disease mortality, 0.73 (95% CI: 0.60, 0.88) for chronic obstructive pulmonary disease mortality, and 0.93 (95% CI: 0.87, 0.99) for cancer mortality; after correction for regression dilution bias, the respective HRs were 0.61 (95% CI: 0.53, 0.70), 0.51 (95% CI: 0.35, 0.73), and 0.84 (95% CI: 0.74, 0.95) (33). These findings emphasize the importance of undertaking repeated dietary assessments over time in cohort studies.

In addition to regression dilution bias, measurement errors in the dietary assessment as well as in covariates can also have an important impact on associations between diet and disease. In univariate analyses, measurement errors tend to attenuate associations between diet and disease when analyzing data from prospective studies; however, in multivariable analyses, measurement errors can both attenuate and exaggerate associations between an exposure and an outcome because measurement errors in the covariates can lead to residual confounding which is difficult to get rid of (227). To date, relatively few dietary studies have made corrections for measurement errors in their analyses. In the EPIC study the HR of ischemic heart disease mortality per 80 g/d of fruit and vegetables was 0.97 (95% CI: 0.95, 0.99) in the uncalibrated analysis, whereas it was 0.95 (95% CI: 0.91, 0.99) in the calibrated analysis (228). For all-cause mortality the uncalibrated and calibrated HRs were 0.97 (95% CI: 0.96, 0.99) and 0.94 (95% CI: 0.91, 0.97), respectively, per 200 g/d of fruit and vegetables (111). Although the differences in these HRs may seem small, this is at least partly due to the small or moderate sizes of the increments of fruit and vegetable intake used. Given that most of the available studies to date neither have used updated measurements of dietary intake nor have corrected for measurement errors, based on the aforementioned results, it seems likely that the observed associations between plant foods and chronic diseases and mortality (31, 52, 68) may be somewhat conservative estimates of the true underlying reduction in risk.

Future Directions

Although there is a growing and impressive body of epidemiological and experimental evidence supporting recommendations for diets high in plant foods (31, 52, 68, 70, 137, 144, 145, 158), a number of gaps have been identified in the current knowledge. Much of the available evidence to date has been on the association between plant foods and risk of type 2 diabetes, coronary artery disease, stroke, cancer, and all-cause mortality. However, increasing amounts of data suggest there may be associations between intake of plant foods and other less common and less investigated causes of death as well (32, 33, 52, 68) and these findings need further examination in future studies. In addition, although we were able to detect associations between some specific types of fruits and vegetables and reduced risk of cardiovascular disease, cancer, and mortality, further studies are needed on specific types of fruits and vegetables, whole grains, and nuts because for many subtypes of plant foods there was only a very limited number of studies published. More studies on subtypes of plant foods are also needed in relation to less common diseases and causes of death. Additional studies are needed on preparation and processing methods, e.g., cooked compared with raw vegetables and salted compared with unsalted nuts. Any additional studies should also report results for more subgroups in order to rule out confounding by smoking, alcohol, adiposity, physical activity, and other dietary factors and use of online supplements could facilitate publication of such results when there is limited space available in journal articles. More studies using biomarkers of fruit and vegetable intake would also be desirable. Recent studies have found specific biomarkers for specific subtypes of fruits and vegetables (229), which might be important for better assessment of dietary intake. The use of online FFQs may make it cheaper and more feasible to collect repeated dietary assessments so dietary changes during follow-up can be taken into account and it would also be desirable if more studies made corrections for regression dilution bias and measurement errors. Because much of the current literature is from North America and Europe with a few studies from Asia, more studies are needed from Asia, the Middle East, Africa, and South America.

Conclusions

Diets high in plant foods including fruits, vegetables, whole grains, and legumes are of major importance for public health because of reductions in the risk of chronic diseases and premature mortality. The current results strongly support dietary recommendations to increase fruit and vegetable intake, but suggest further benefits beyond the currently recommended 5-a-day up to an intake of 800 g/d, particularly for coronary artery disease and for mortality from stroke. Studies using antioxidant biomarkers of fruit and vegetable intake suggest more linear associations than studies using dietary questionnaires. The current results also support recommendations to increase whole-grain intake to 225 g/d and nut intake to 15–20 g/d. Given the lack of quantitative dietary recommendations regarding whole-grain and nut intake in many countries, these results provide the best available evidence currently. Diets high in plant foods could potentially prevent several million premature deaths each year if adopted globally.

Notes

Published in a supplement to Advances in Nutrition. This supplement was sponsored by the Harding-Buller Foundation of Ohio. The contents are solely the responsibility of the authors and do not necessarily represent the official views of the sponsors. Publication costs for this supplement were defrayed in part by the payment of page charges. The opinions expressed in this publication are those of the authors and are not attributable to the sponsors or the publisher, Editor, or Editorial Board of Advances in Nutrition.

Supported by the School of Public Health, Imperial College London and the South-East Regional Health Authorities of Norway.

Author disclosures: The author report no conflicts of interest.