Abstract

BACKGROUND: Several systematic reviews were published in the last years, investigating vegetables & fruits in relation to CHD & stroke (43-46). Conclusions about possible associations differed between the reviews. These differences were due to variations in inclusion/exclusion criteria for articles. Until now, no review was published examining interactions with possible effect modifiers. And none of the reviews examined the relation with subgroups of vegetables/fruits, or specific vegetable & fruit items.
OBJECTIVES:

• To review all prospective studies which provided data about total fruits or vegetables, their subgroups, and specific types of vegetables and fruits, in relation with cardiovascular disease, heart disease, or stroke risk and mortality/survival.
• To define the amount of consumption found to be related with the described effects on CVD.
• To define possible effect modification by confounders.

DATA SOURCE: The Pubmed database was searched (No start date - Jul 6, 2009) for relevant articles using the keywords "prospective, cohort, follow-up, longitudinal" combined with "vegetables, fruit", and a fair amount of other keywords. The exact search term is described in the methods.
Prospective studies published in the English language were included. Reference lists were searched for additional articles.
RESULTS: 85 articles were found which provided information about 43 different cohorts. Of these, 0 articles were excluded.
Results about the relation with total vegetables & total fruits are briefly described. Evidence for other associations is described in the conclusion. Disease incidence and death were considered different end points.

• Total CVD risk.
  -Total vegetables: Data was provided by 6 cohorts, including 8,744 cases. Total vegetables possibly protects against CVD risk (- 7%) among non-users of multivitamin supplements, and among low carbohydrate consumers (≤ 40 en%).
  -Total fruit: Data was provided by 6 cohorts, including 6,000 cases. Total fruit possibly protects against CVD risk (- 13%). This effect may be restricted to subjects with moderate carbohydrate consumption (40-55 en%).
• Heart disease risk.
  -Total vegetables: Data was provided by 8 cohorts, including 4,502 cases. Total vegetables possibly protect against CHD risk (- 19%). This effect may be restricted to non-users of mulvititamin supplements, and current smokers.
  -Total fruit: Data was provided by 9 cohorts, including 4,520 cases. Suggestive evidence was found for a protective effect among men (- 18%).
• Stroke risk.
  -Total vegetables or total fruit: Data about total stroke risk was provided by 4 cohorts, including 663 cases. Inconclusive evidence was found for an association.
  -Total fruit: Data about ischemic stroke risk was provided by 5 cohorts, including 3,647 cases. Total fruit possibly protects against ischemic stroke risk (- 22%).
• Total CVD death.
  -Total vegetables: Data was provided by 5 cohorts, including 3,801 cases. Inconclusive evidence was found for an association.
  -Total fruit: Data was provided by 6 cohorts, including 5,003 cases. Suggestive evidence was found for a protective effect (- 15%).
• Heart disease death.
  -Total vegetables: Data was provided by 9 cohorts, including 2,629 cases. Suggestive evidence was found for a protective effect (- 23%).
  -Total fruit: Data was provided by 13 cohorts, including 3,628 cases. Inconclusive evidence was found for an association.
• Stroke death.
  -Total vegetable: Data about total stroke mortality was provided by 3 cohorts, including 1,381 cases. Inconclusive evidence was found for an association.
  -Total fruit: Data was provided by 5 cohorts, including 3,465 cases. Total fruit possibly protects against total stroke mortality (- 27%).

CONCLUSION: Protective effects of both total vegetables, and total fruit consumption were found against most CVD/CHD end points. Fruits, but not vegetables possibly protect against ischemic stroke risk and total stroke mortality. Protective effects of both citrus fruits, and green leafy vegetables were found against all disease risk end points. Protective effects of vitamin-C rich vegetables & fruits were found against CVD risk, and protective effects of cruciferous vegetables were found against both CHD risk and stroke risk. Very little evidence was found for protective effects of specific vegetables or fruits: citrus fruit juice against ischemic stroke risk; both apples and broccoli against CHD mortality among women; and pickles against stroke mortality.
All levels of evidence were low (suggestive-possible). And effects were often restricted to subgroups of the population (e.g., non-users of multivitamin supplements, low-moderate consumers of carbohydrates [≤ 55 en%], or current smokers/nonsmokers).
PERSPECTIVE: Because few effects were published in servings/grams/cups over a given time period, and in increasing units of consumption, no level of consumption could be calculated for the associations. This given fact hinders evidence based health recommendations.
Most of the evidence was based - to a large extend - on findings from 2 cohorts: "The Nurses' Health Study" & "The Health Professionals' Follow-Up Study". Without these 2 cohorts, most of the evidence would disappear.
A lot of cohorts are currently providing information about diet & various health outcomes. Some cohorts are of very large size (200,000 to 1,000,000 subjects). In the coming years it can be expected that within each of these large cohorts more than 10,000 to 50,000 subjects will have developed CVD. Since the evidence relating vegetables or fruits to total CVD risk is currently based on a cumulative amount of cases less than 9,000, it is not hard to understand that current evidence could change in any direction in the near future.
AUTHOR'S NOTE: Information was presented about the relation with total CVD, heart disease, and stroke. And information was further stratified into risk, or death. The choice could be made to group data together (risk & death, or total CVD & CHD & stroke). However, none of these options would result in probable/convincing evidence for an effect of any of the dietary variables according to the methodology used.

Introduction:

Several systematic reviews with meta-analysis were published in the last years, investigating vegetables & fruits in relation to CHD & stroke. Results from these reviews will be discussed briefly:

• Dauchet L (43) Stroke. Included 4 cohorts for vegetables and 5 for fruits. A significant protective effect was found of total fruits (RR = 0.89; 95% CI = 0.85-0.93), but no association was found with total vegetables (RR = 0.97; 95% CI = 0.92-1.04) for each additional portion/day.
• He FJ (45) Stroke. Included 6 cohorts for vegetables and 6 cohorts for fruits. Significant protective effects were found of both total fruits (RR = 0.72; 95% CI = 0.66-0.79), and total vegetables (RR = 0.81; 95% CI = 0.72-0.90) for > 5 vs < 2 servings/day.
• Dauchet L (44) CHD. Included 7 cohorts for vegetables and 6 for fruits. Significant protective effects were found of both total fruits (RR = 0.93; 95% CI = 0.89-0.96), and total vegetables (RR = 0.89; 95% CI = 0.83-0.95) for each additional portion/day.
• He FJ (46) CHD. Included 9 cohorts for vegetables and 9 cohorts for fruits. Significant protective effects were found of both total fruits (RR = 0.87; 95% CI = 0.80-0.95), and total vegetables (RR = 0.84; 95% CI = 0.76-0.92) for > 5 vs < 3 servings/day.

Excluded articles: In all 4 meta-analysis', exclusion criteria were used. Dauchet L (43) excluded studies that did not use a validated dietary assessment tool, Dauchet L (44) excluded studies that used frequency of consumption as dietary assessment, and He FJ (45, 46) excluded studies that used only two categories of fruits and vegetable intake (e.g., daily vs never). Though these exclusion criteria may be justified for their specific types of analysis, any effects from studies which are excluded from the analysis could have possibly changed the effect size from the final results.
Differences between meta-analysis': Effect sizes differed between the results for both vegetables and fruits. These are due to the way results were presented. Dauchet L (43, 44) presented results as a continuous variable (per increment of 1 portion/day), whereas He FJ (45, 46) presented results as a categorized variable (high vs low consumption).
Also, different inclusion criteria led to different studies in the analysis. E.g., for the analysis regarding stroke and vegetables, He FJ (45 included an article in which green-yellow vegetables consumption was examined instead of total vegetables (Sauvaget C [28]), and also included 2 cohorts (Keli SO [6], Hirvonen T [13]) which were not included by the analysis from Dauchet L (43). On the other hand, Dauchet L (43) included an article (Gillman MW [8]) which was not included by the analysis from He FJ (45).
Missing articles: For the relation between stroke and total vegetables, one article was not included by both analysis' (Yokoyama T [3]). In addition, one missing article mentioned that "no significant association was found between fresh vegetables and stroke mortality" (Ross RK [16]).
He FJ (45) apparently included studies examining the relation with green-yellow vegetables. According to this criterium, the author missed one article which would have an enormous impact on the results: Kinjo Y (5) examined the relation between green-yellow vegetables consumption and total stroke mortality in a cohort of very large size, and after 15 years of follow-up, 11,030 subjects died of stroke. The RR from this cohort was 1.03 (0.90-1.17) for the highest vs lowest tertile of consumption. If results from this cohort were to be included in the analysis by He FJ (45), this would strongly attenuate the relation. Though no meta-analysis was performed, the effect size can be estimated from adding up the RR's after adjustment for the amounts of cases (See table 1).

Summary: Different methodological analysis' can have a large impact on the results, and thereby on the conclusion. Even when 2 authors create a meta-analysis about the same variables. Also, missing one single article from the literature review, or excluding an article for any reason can have a large impact on the results and practically eliminate an association.

Methods

The methodology for the systematic review can be found in the menu. Contrary to the systematic reviews mentioned in the introduction, no articles were excluded from the analysis.
All former systematic reviews only examined total vegetables and total fruits in relation to the different CVD end points. This systematic review will be the first to examine the relation with subgroups of vegetables and fruits, and specific vegetables and fruits. Also, this systematic review will be the first to examine the interaction with effect modifiers, other than dietary assessment method. After all, effects may only be found among certain subgroups of the population.

Total vegetables and total cardiovascular disease.

CVD risk: Data about total CVD risk was provided by 6 cohorts, including a total of 8,744 cases.
A significant protective effect was found from pooling results of 2 cohorts of moderate-large size (Hung HC. 2004 [10]), and including 3,634 cases. No other associations were found. The average RR = 0.96 (excluding incomplete data from Panagiotakos D [42]).
The inclusion of results from one cohort is debatable (Yen AM. 2008 [40]). The KCIS Study included 21,906 subjects, 3,163 of whom were diagnosed with CVD within < 3 years. This was because the definition of CVD was much wider than among the other cohorts. Exclusion of results from this cohort somewhat strengthened the effect size: Average RR = 0.93.
Effect modification: The protective effect from the pooled results of 2 cohorts was restricted to non-multivitamin supplement users (Hung HC. 2004 [10]), and was also restricted to subjects with low carbohydrate consumption (< 40 en%; Joshipura KJ. 2009 [10]. See extended table).
CVD mortality: Data about total CVD mortality was provided by 5 cohorts, including a total of 3,801 cases.
Protective effects were found in 2 cohorts, though no data was available about the strength or significance of the association in one cohort (Vollset SE [2]), and the effect was restricted to women in the other cohort (Nakamura K [24]). The average RR = 0.91 (excluding incomplete data from Vollset SE [2], and Chang-Claude J [33]).

Conclusion: Inconclusive evidence was found for an association between total vegetables consumption and total CVD risk or CVD death. Exclusion of one cohort with debatable results changed the evidence for total CVD risk: Total vegetables possibly protects against total CVD risk (- 7%). But this evidence was restricted to non-users of multivitamin supplements, and to low carbohydrate consumers (< 40 en%). No evidence was found for an association with total CVD death.

Total vegetables and heart disease.

Background: In 2007, He FJ. published a meta-analysis of 9 cohort studies relating vegetables to CHD risk. This analysis excluded data from "The Shibata Study" (Tanaka H [3]), and "The Study Of Men Born In 1913" (Strandhagen E [19]). But it included data from 3 other cohorts which are not added to the following tables, because the associaton with total vegetables was not examined. Instead, specific subgroups were examined. Information about these subgroups can be found in the menu. Excluded articles are as follows:

• Fraser GE (1992 [4]). The Adventist Health Study provided information about salads and legumes, but not total vegetables.
• Mann JI (15). Provided information about green vegetables, but not total vegetables.
• Liu S (22). Provided information about vegetables rich in carotenoids, but not total vegetables.

CHD risk: Data about heart disease risk was provided by 8 cohorts, including 4,502 + X cases. No amount of cases was provided by one cohort (Tanaka H [3]).
Significant protective effects were found in 3 cohorts (Joshipura KJ [10] among women; Hirvonen T [13]; and Buckland G [26] among men) including 2.729 cases (61% of all cases) No other associations were found. The average RR = 0.81 (excluding incomplete data from Tanaka H [3], and Strandhagen E [19]).
Effect modification: The protective effect from the pooled results of 2 cohorts was restricted to non-multivitamin supplement users (Hung HC. 2004 [10]). In this same analysis, the effect was significant among current smokers only. And a significant protective effect was also found in a cohort examining current smokers only (Hirvonen T [13]).
CHD mortality: Data about heart disease mortality was provided by 9 cohorts, including 2,629 cases.
Significant protective effects were found in 3 cohorts (Sahyoun NR [11], Hirvonen T [13], and Tucker KL [31]), including 987 cases (38% of all cases). No other associations were found. The average RR = 0.77 (excluding Knekt P [7] because no amounts of cases were provided stratified by gender. And excluding Strandhagen E [19], Albert CM [22], and Chang-Claude J [33] because of incomplete data).

Conclusion: Significant protective effects of total vegetables against CHD risk were found in 3 cohorts of moderate-large size, and including 61% of all cases. Total vegetables consumption possibly protects against CHD risk (- 19%). This effect may be restricted to non-users of multivitamin supplements, and current smokers.
Significant protective effects of total vegetables against CHD death were found in 3 cohorts including 38% of all cases. Suggestive evidence was found for a protective effect of total vegetables against CVD death (- 23%).

Total vegetables and stroke.

Background: In 2006, He FJ. published a meta-analysis of 6 cohort studies relating vegetables to stroke risk. This analysis excluded data from "The Shibata Study" (Tokoyama T [3]), and "The Framingham Study" (Gillman MW [8]), but it included data from 1 other cohort which was not added to the following tables.
The excluded article used data from "The Hiroshima Nagasaki Life Span Study" (Sauvaget C [28]). This study provided information about about green-yellow vegetables, but not total vegetables. Therefore, information about this variable is added to the related item in the menu.

Stroke risk: Data about total stroke risk was provided by 4 cohorts, including 663 cases.
A significant protective effect was found in one cohort of very small size (Tokoyama T [3]). No other associations were found. The average RR = 0.77.
Data about stroke subtypes was available from 3 other cohorts (Joshipura KJ [10], Larsson SC [13], Johnsen SP [26]). Effect sizes were similar to those for total stroke risk (RR = 0.62 to 1.00. See extended table).
Stroke mortality: Data about total stroke mortality was provided by 3 cohorts, including 1,381 cases.
A significant protective effect was found in one cohort of very small size (Ness AR [32]). No other associations were found. The average RR = 1.04.

Conclusion: Few associations were found. Inconclusive evidence was found for an association between total vegetables and total stroke risk or total stroke mortality. Though the effect size from the average RR for stroke risk does not exclude the possibility of an association.

Total fruit and total cardiovascular disease.

CVD risk: Data about total CVD risk was provided by 6 cohorts, including 6,000 cases.
A significant protective effect was found from pooling results of 2 cohorts of moderate-large size (Hung HC. 2004 [10]), and in 2 additional cohorts (Cox BD [20], Takachi R [36]), though the effect was restricted to women in 2 cohorts (Cox BD [20], Takachi R [36]). This analysis included 4,377 cases (73% of all cases). No other associations were found. The average RR = 0.87 (excluding incomplete data from Panagiotakos D [42]).
Effect modification: The protective effect from the pooled results of 2 cohorts was restricted to past- and current smokers (Hung HC. 2004 [10]), and was also restricted to subjects with moderate carbohydrate consumption (40-55 en%; Joshipura KJ. 2009 [10]. See extended table).
On the contrary, the effect was restricted to nonsmokers in one other cohort (Takachi R [36]).
CVD mortality: Data about total CVD mortality was provided by 6 cohorts, including 5,003 cases.
Significant protective effects were found in 2 cohorts (Nöthlings U [26], Nagura J [37]), including 2.760 cases (55% of all cases). A protective effect was found in a third cohort, but no data was provided about the strength or significance of the association (Vollset SE [2]). In a fourth cohort, a significant protective effect was found among women, but not men (Appleby PN [12]), but the RR for the whole cohort was include in the anlysis, because this RR was better adjusted for confounders. No other associations were found. The average RR = 0.85 (excluding incomplete data from Vollset SE [2], and Chang-Claude J [33]).
CVD risk & CVD mortality: When CVD risk and mortality end points were combined, significant protective effects were found among women, but not men in 3 cohorts. Therefore, a stratified analysis was conducted to examine the possible difference in effect between both genders.
Data about men was provided by 5 cohorts, including 3,496 cases. No associations were found. The average RR = 0.89.
Data about women was provided by 6 cohorts, including 3,920 cases. Significant protective effects were found in 3 cohorts (Appleby PN [12], Cox BD [20], Takachi R [36]). incuding 1.354 cases (35% of all cases). And all other RR's were < 1. The average RR = 0.85.

Conclusion: Significant protective effects of total fruits against total CVD risk were found in 4 cohorts, 3 of which were of moderate-large size. Total fruits consumption possibly protects against total CVD risk (- 13%). This effect may be restricted to subjects with moderate carbohydrate consumption (40-55 en%).
Protective effects of total fruits against total CVD death were found in 3 cohorts, one of which was of moderate size. Suggestive evidence was found for a protective effect of total fruit consumption against total CVD death (- 15%).
Analysis stratified by gender showed suggestive evidence for a protective effect among women when disease risk & death were considered one end point (- 15%). While no associations were found among men. This finding should be interpreted with care, because the RR's between men and women differed little.

Total fruit and heart disease.

Background: In 2007, He FJ. publised a meta-analysis of 9 cohort studies relating fruit to CHD risk. This analysis excluded data from "The Shibata Study" (Tanaka H [3]), and "The Study Of Men Born In 1913" (Strandhagen E [19]). But it included data from 2 other cohorts which are not added to the following tables, because the association with total fruits was not examined. Instead, specific subgroups were examined. Information about these subgroups can be found in the menu. Excluded articles are as follows:

• Knekt P (7). Provided information about fruit other than apples & berries, instead of total fruit.
• Liu S (22). No information about fruit is published in the article referred to by the meta-analysis. It should be noted that He FJ. possibly had access to information about this cohort which is not yet published.

CHD risk: Data about heart disease risk was provided by 9 cohorts, including 4,520 + X cases (No amount of cases was provided by one cohort (Tanaka H [3]).
Significant protective effects were found in 2 cohorts (Joshipura KJ [10] among men; Dauchet L [30]), including 1,196 cases (26% of all cases). No other associations were found. The average RR = 0.84 (excluding incomplete data from Tanaka H [3], and Strandhagen E [19]). Since the protective effects were found among men only, a separate analysis was made, stratified by gender:
Data for men was provided by 4 cohorts (10, 13, 26, 30), including 2,798 cases. Significant protective effects were found in 2 cohorts, including 1,196 cases (43% of all cases). The average RR = 0.82.
Data for women was provided by 3 cohorts (10, 21, 26), including 1,379 cases. No associations were found, but the average RR was similar to the one for men: RR = 0.84.
CHD mortality: Data about heart disease mortality was provided by 13 cohorts, including 3,628 cases.
Significant protective effects were found in 2 cohorts, including 831 cases (Appleby PN [12], Strandhagen E [19]). In addition, nonsignificant protecive effects were found in 2 other cohorts (Knekt P [7], Nagura J [37]) though this association was restricted to women in one cohort (Knekt P [7]). No other associations were found. The average RR = 0.90 (excluding incomplete data from Strandhagen E [19], Albert CM [22], Chang-Claude J [33]).

Conclusion: Significant protective effects of total fruits against CHD risk were found in 2 cohorts of small-moderate size. Both cohorts examined men only. Suggestive evidence was found for a protective effect of total fruit against CHD risk among men only (- 18%), though the effect size did not differ between men and women.
Some protective effects of total fruits against CHD death were found, but these were not always significant. Inconclusive evidence was found for an association between total fruit consumption and CHD death.

Total fruit and stroke.

Total stroke risk: Data about total stroke risk was provided by 4 cohorts, including 663 cases. No associations were found. The average RR = 0.88.
Ischemic stroke risk: Data about ischemic stroke risk was provided by 5 cohorts, including 3.647 cases. Significant protective effects were found from pooled results of 2 cohorts (Joshipura KJ [10]), and in 2 other cohorts (Larsson SC [13], Johnsen SP [26]). The RR was well below 1 in the remaining cohort (Tokoyama T [3]). The average RR = 0.78.
Total stroke mortality: Data about total stroke mortality was provided by 5 cohorts, including 3,465 cases. Significant protective effects were found in both Japanese cohorts (Sauvaget C [28], Nagura J [37]), including 2.781 cases (80% of all cases). No other associations were found, but all RR's were well below 1. The average RR = 0.73 (excluding incomplete data from Ross RK [16]).

Conclusion: No associations were found between total fruits and total stroke risk. But significant protective effects were found in 4 out of 5 studies examining ischemic stroke risk as the end point. Total fruit consumption possibly protects against ischemic stroke risk (- 22%).
Significant protective effects of total fruits against total stroke death were found in 2 cohorts of moderate size, and including 80% of all cases. Total fruits consumption possibly protects against total stroke death (- 27%).

Total fruits & vegetables

Some articles did not provide information about either "total vegetables" or "total fruit", but looked at their combined effect.

Information about this combined effect is only presented if no information about either "total vegetables" or "total fruits" was provided from a cohort.
Data from "The Nurses' Health Study" can be found in the extended version of the table (10). This data was not included in the following analysis, because stratified data about total vegetables & total fruits is available from this cohort. Still, 2 articles about this cohort were published in which this stratification was not made.

Total cardiovascular disease: No data about the association with total CVD risk was found.
Data about CVD death was provided by 3 cohorts of (very) small size, including a total of 1,638 cases.
Significant protective effects were found in 2 cohorts (Bazzano LA [23], Rissanen TH [25]). A nonsignificant protective effect among never smokers was found in the third cohort (Genkinger JM [29]).

Coronary heart disease: Data about heart disease risk was provided by 6 cohorts of (very) small size, and including a total of 3,756 + X cases (no amount of cases was available from Gillman MW [8]).
A significant protective effect was found in 2 cohorts (Bingham S [26], Holmberg S [41]). But no associations were found in the remaining cohorts.
Data about IHD mortality was provided by 1 cohort of small size (Bazzano LA [23]). A nonsignificant protective effect was found.

Stroke: Data about stroke risk was provided by 2 cohorts of small size. A significant protective effect was found in 1 cohort (Bazzano LA [23]), but no association was found in the other cohort.
Data about stroke mortality was provided by 1 cohort of small size (Bazzano LA [23]). A significant protective effect was found.

Conclusion: Cohort sizes were small, and associations cannot be attributed to either vegetables or fruits. Nonetheless, these findings show few association with disease risk, but strenghten the possibility of a protective association against CVD mortality.