Abstract

BACKGROUND: In 2007, the World Cancer Research Fund (WCRF) published a systematic review with information about a large amount of life style variables in relation to a large variety of cancer types. The WCRF concluded that there was a) limited evidence suggesting that non-starchy vegetables protect against lung cancer, and b) convincing evidence suggesting that fruits protect against lung cancer. However, little data was presented about specific vegetable and fruit subgroups or items in the main report, and no data was presented about possible effect modification by confounders.
OBJECTIVES:

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

DATA SOURCE: The Pubmed database was searched (No start date - Feb 10, 2009) for relevant articles using the keywords "cancer, neoplasm, or neoplasms" combined with "vegetables, vegetable, fruit, or fruits" 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: 61 articles were found which provided information about 31 different cohorts. Of these, 0 articles were excluded.
Results are described when any evidence for an association - as defined in the Methods - was found. In addition, data about total vegetables or fruits is described.
Total disease risk.

• Total vegetables: 32 articles providing information about 20 different cohorts were found. Total vegetables possibly protect against lung cancer risk among men (RR = 0.82). This effect was found at the level of consumption of ≥ 170 g/day. No association was found among women (RR = 1.01). Suggestive evidence was found for a protective effect among current smokers (RR = 0.85), while no evidence was found for an association among former-, and never smokers. The protective effect may be restricted to cancer cases diagnosed at age < 65.
• Total fruits: 34 articles, providing information about 20 different cohorts were found. Total fruits probably protect against lung cancer risk at consumption of ≥ 127 g/day (RR = 0.85). Effect sizes showed no evidence for effect modification by gender. Suggestive evidence was found for a protective effect among current smokers (RR = 0.86), while no evidence was found for an association among former-, and never smokers.
• Suggestive evidence for a protective effect of the following variables was found: lettuce at ≥ 1.26 servings/week, tomatoes at ≥ 0.5 tomato/day (RR = 0.83), green leafy vegetables at ≥ 0.5 serving/day (RR = 0.90), and cruciferous vegetables (RR = 0.94).
• Possible protective effects of the following items were found: apples/pears at > 1 apple or pear/week (RR = 0.78), oranges/tangerines at > 1 orange or tangerine/week (RR = 0.74) and tomato sauce (no level of consumption could be defined).
• Possible protective effects at intermediate levels of consumption were found of carrots (0.5 to < 1 serving/week), and - among women - of citrus fruit at 7 servings/week (RR = 0.86).
• Orange/grapefruit juice probably protects against lung cancer risk at 24-55 g/day (RR = 0.81), and possibly protects against this disease at higher consumption.

Advanced stage/metastatic disease risk or disease progression.
No data was found.
Mortality risk.

• Total vegetables: Data about 4 cohorts was found. No association was found in any cohort.
• Total fruits: 11 articles providing information about 9 different cohorts were found. Total fruit probably protects against lung cancer mortality. Stratified by sex, fruits were probable protective among men (RR = 0.60), while little data is available about women. The level of consumption for this effect was found at (almost) daily consumption.
• Oranges possibly protect against lung cancer mortality among men at > 3 oranges/wk (RR = 0.73).
• Suggestive evidence was found for a protective effect of citrus fruit among men (RR = 0.76). No level of consumption could be defined for this effect.

CONCLUSION:
Total vegetables possibly protect against lung cancer risk among men (≥ 170 g/day), but not women. This protective effect may be restricted to current smokers, and cancer cases diagnosed at age < 65 years. Total fruits probably protect against lung cancer risk (≥ 127 g/day) among men & women. This effect may also be restricted to current smokers. Suggestive evidence for a protective effect was found of green leafy vegetables and cruciferous vegetables, and among women, citrus fruit is possibly protective (median 7 servings/wk).
Among specific vegetable or fruit items, apples/pears (> apple or pear/wk), oranges/tangerines (> 1 orange or tangerine/wk), tomato sauce and an intermediate level of carrot consumption (0.5-< 1 serving/wk) possibly protect against lung cancer risk. Orange/grapefruit juice probably protects against lung cancer risk at an intermediate level of consumption (24-55 g/day), and a possible protective effect was found with higher consumption. Suggestive evidence was found for a protective effect of lettuce (≥ 1.26 servings/week), and tomatoes (≥ 0.5 tomato/day). For lung cancer mortality, no evidence was found for an effect of vegetables. Among men, total fruits are probably protective at (almost) daily consumption, oranges are possibly protective (> 3 oranges/wk), and suggestive evidence was found for a protective effect of citrus fruit.
PERSPECTIVE: Though evidence for protective effects of both total vegetables & total fruits could already be found at consumption levels just below the recommended levels of consumption, effect sizes were just as strong for specific vegetable or fruit items at low levels of consumption: Apples, oranges, and tangerines were just as protective at consumption of ≥ 1 item/week, as were high consumption of total fruits, or vegetables.
LIMITATIONS: Though this review suggests a probable protective effect of fruit against lung cancer mortality, it should be noted that in most cohorts consumption was measured prior to cancer diagnosis. Little is known about the effect of fruit consumption following cancer diagnosis.

Introduction.

Results from the WCRF: In 2007, the WCRF published a systematic review with information about a large amount of life style variables in relation to a large variety of cancer types (World Cancer Research Fund). The WCRF concluded that there was a) limited evidence suggesting that non-starchy vegetables protect against lung cancer, and b) convincing evidence suggesting that fruits protect against lung cancer. Meta-analysis showed RR's of 0.95 (0.92-0.98), and 0.94 (0.90-0.97) - per 80 g serving/day - for total vegetables, and total fruits, respectively.
Continuous vs categorized analysis: The WCRF chose to base it's evidence on results presented as a continuous analysis. Calculations showed that the risk of lung cancer would decrease by 5, and 6% for each 80 g/serving of total vegetables, and total fruits, respectively. This type of calculation suggests that a) results from prospective studies show a straight line, and that b) an additional benefit can be found with higher intake, even among subjects already having high consumption levels. The WCRF results also suggest that c) subjects consuming 480 g of vegetables or fruits would have 25, and 30% lower risks of lung cancer, compared with subjects consuming 80/day, suggesting a strong effect size at much higher consumption levels.
However, most articles published results as a categorized variable, meaning that effects are shown for specific levels of consumption. Results in my systematic review will be presented as a categorized variable if available from the related articles. And this analysis will show that there is no evidence for any of the effects as they are suggested to be by the WCRF:

• Pooled analysis shows that effect sizes among the 3rd to 5th quintile of consumption are similar for both total vegetables & total fruits (Smith-Warner SA [24]). My own analysis will also show that there is more evidence for a tresshold effect - or for a curved line - than for a continuous effect.
• Categorized analysis shows effect sizes of high vs low consumption levels are weaker than would be expected from the continuous analysis.

Reference:
World Cancer Research Fund / American Institute for Cancer Research. Food, Nutrition, Physical Activity, and the Prevention of Cancer: a Global Perspective. Washington DC: AICR, 2007. Full text

Total vegetables and lung cancer risk.

About the male analysis: In 2003 a pooled analysis of 8 prospective studies was published (Smith-Warner SA [24]). The analysis included data from a total of 1,808 men of which 298 were from The ATBC Study. In 2002 another publication about The ATBC Study included data from 1,644 men (Holick CN [18]), which is nearly as much as the total amount of cases from the pooled analysis. A similar finding was done with The Netherlands Cohort Study. For this systematic review it was chosen not to include data about men from the pooled analysis as a whole, but from the individual cohorts, because this allowed for a much larger amount of cases in the analysis.
As a consequence of this decision, data about men from The New York State Cohort (including 392 male cases) can not be included in these specific parts of the review, since no article is published about this specific cohort.

Results:

• Men & women combined: 32 articles, providing information about 20 different cohorts were found, including 14,125 cases. Significant protective effects were found in 3 cohorts among men (Balder HF [11], Holick CN [18], George SM [29]) including 7,162 male cases. The average RR = 0.90 (excluding incomplete data from Shekelle RB [1], and Ratnasinghe D [15]).
• Men: Data about 11 cohorts including a total of 8,107 cases was found. A significant protective effect was found in 3 cohorts (Balder HF [11], Holick CN [18], George SM [29]) including 7,162 cases (88% of all cases). No associations were found in the remaining 8 cohorts.
  The average RR = 0.82 (excluding incomplete data from Shekelle RB [1], and Ratnasinghe D [15]).
• Women: Data about 9 cohorts including a total of 3,863 cases was found. No significant associations were found. The average RR = 1.01
• In three other cohorts including a total of 2,294 cases, risk was not stratified by sex. No associations were found. The average RR = 0.98

Inclusion of intermediate levels of consumption:
(Non)significant effects at any level of consumption among men were as follows:

• Balder HF (11) At 294 g/day (highest quintile of consumption).
• Holick CN (18) At ≥ 110 g/day.
• George SM (29) At 170-235, and at ≥ 424 g/day (2nd and 5th quintile of consumption).

Among men, overlapping effects were found at the level of ≥ 170 g/day.

(Non)significant effects at any level of consumption among women were as follows:

• Smith-Warner SA [24] At the 3rd quintile of consumption. No level of consumption could be defined, because the pooled analysis included study-specific quintiles of consumption from the 6 cohorts.

Among women, the possibility of an association at an intermediate level of consumption can not be excluded, but is impossible to assess based on this data.

Effect modification by variables other than sex and smoking status:
A significant protective effect among cancer cases diagnosed at age < 65 years, but not ≥ 65 years was found in a pooled analysis of 8 cohorts (Smith-Warner [24]). In the same analysis, no evidence of effect modification was found by the number of fruit and vegetable questions included on a study's FFQ.
No evidence of effect modification was found by alcohol intake or multivitamin use in two cohorts (Feskanich D [14]), or by use of beta carotene/vitamin A supplements in another cohort (Neuhouser ML [22]).

Conclusion: For men, a significant protective effect was found in 3 cohorts of moderate-very large size, and no (non)significantly increased risks were found in the remaining cohorts. Total vegetables possibly protect against lung cancer risk among men (- 18%). Levels of consumption for this effect were ≥ 170 g/day, and the effect may be restricted to cancer cases diagnosed at age < 65 years. No associations were found among women.

Total vegetables and lung cancer risk. Stratified by smoking status.

Background:
About the analysis: For this systematic review, data is stratified by smoking status into current smokers, former smokers, and never smokers.
The JPHC Study (Liu Y [25]) and CARET Study (Neuhouser M [22]) provided data about "ever smokers" without stratifying into former vs current smokers. Likewise, The Finnish Mobile clinic Health Cohort (Knekt P [8]) provided data about "current nonsmokers" without stratifying into nonsmokers vs former smokers. It was chosen not to include this data in the following tables.
About the analysis of current smokers: In 2003 a pooled analysis of 8 prospective studies was published (Smith-Warner SA [24]). The analysis included data from a total of 1,915 smoking cases of which 298 were from The ATBC Study. In 2002 another publication about The ATBC Study included data from 1,644 smoking cases (Holick CN [18]), which is nearly as much as the total amount of cases from the pooled analysis. For the systematic review it was chosen not to include data about current smokers from the pooled analysis as a whole, but from the individual cohorts, because this allowed for a much larger amount of cases in the analysis.

Results:

• Current smokers: Data about current smokers was provided by 8 cohorts, including 6,206 cases. Significant protective effects were found in 3 cohorts (Voorrips LE [11]), Feskanich D [14] among women, and Holick CN [18]), including 2,436 cases (39% of all cases). And a nonsignificant protective effect was found in a fourth cohort (Steinmetz KA [10]). No other (non)significant associations were found. The average RR = 0.85
• Former smokers: Data about former smokers was provided by a pooled analysis of 7 cohorts, and 2 additional cohorts which were both of very large size. The analysis included 4,217 cases. A significant protective effects was found in one cohort among men, while among women a nonsignificantly increased risk was found (Wright ME [29]). No other associations were found. The average RR = 1.01
• Never smokers: Data about never smokers was provided by a pooled analysis of 7 cohorts, and 3 additional cohorts, including 774 cases. No (non)significant effects were found. The average RR = 0.94

Conclusion: Significant protective effects among current smokers were found in 3 cohorts of moderate-large size, including 39% of all cases. Suggestive evidence was found for a protective effect of total vegetables against lung cancer risk among current smokers (- 15%). Few associations were found among former-, or never smokers, and no evidence was found for an association among these subgroups.

Total vegetables and lung cancer mortality.

Data about lung cancer mortality is provided by 4 cohorts. Three of these cohorts measured food intake prior to cancer diagnosis (5, 9, 16), and another one following diagnosis (19). No associations were found. The average RR = 1.02 (excluding incomplete data from Skuladottir H [19]).

Conclusion: No associations were found. No evidence was found for an association between total vegetables and lung cancer mortality.

Total fruit and lung cancer risk.

About the male analysis: In 2003 a pooled analysis of 8 prospective studies was published (Smith-Warner SA [24]). The analysis included data from a total of 1,808 men of which 298 were from The ATBC Study. In 2002 another publication about The ATBC Study included data from 1,644 men (Holick CN [18]), which is nearly as much as the total amount of cases from the pooled analysis. A similar finding was done with The Netherlands Cohort Study. For this systematic review it was chosen not to include data about men from the pooled analysis as a whole, but from the individual cohorts, because this allowed for a much larger amount of cases in the analysis.
As a consequence of this decision, data about men from The New York State Cohort (including 392 male cases) can not be included in these specific parts of the review, since no article is published about this specific cohort.

Results:

• Men & women combined: 34 articles, providing information about 20 different cohorts were found, including 14,323 cases. Significant protective effects were found in 6 cohorts (Knekt P [8], Balder HF [11], Holick CN [18], Linseisen J [19], Neuhouser ML [22], George SM [29]), 2 of which were of very large size (19, 29). These 6 cohorts included 8.752 cases (61% of all cases). In addition, nonsignificant protective effects were found in a pooled analysis of 6 cohorts (Smith-Warner SA [24]), an in one additional cohort (Alavanja MC [27]), including another 1,445 cases (10% of all cases). No other (non)significant associations were found.
  The average RR = 0.85 (excluding incomplete data from Shekelle RB [1], Ratnasinghe D [15]).
• Men: Data was provided about 11 cohorts including a total of 8,116 cases. A significant protective association was found in 4 cohorts (Knekt P [8], Balder HF [11], Holick CN [18], George SM [29]) including 7,300 cases (90% of all cases), and a nonsignificant protective effect was found in one other study (Shekelle RB [1]). No associations were found in the remaining 6 cohorts).
  The average RR = 0.87 (excluding incomplete data from Shekelle RB [1], Ratnasinghe D [15]).
• Women: Data was provided about 9 cohorts including a total of 3,862 cases. A nonsignificant protective effect was found in a pooled analysis of 6 cohorts (Smith-Warner SA [24]), and a nonsignificant protective effect was found in one other cohort (Alavanja MC [27]) including 1,445 cases. No other associations were found, but all RRs were < 1.
  The average RR = 0.86
• In 4 cohorts including a total of 2,345 cases, risk was not stratified by sex. Significant protective effects were found in 1 cohort (Linseisen J [19]), and in the placebo arm of another cohort (Neuhouser ML [22]), including 1,452 cases. No other associations were found. The average RR = 0.80

Inclusion of intermediate levels of consumption:
Though a fair amount of (non) significant associations were found, sometimes the trend, but not the effect (RR) was (non)significantly differend from 1.00.
(Non)significant effects at any level of consumption among men were as follows:

• Knekt P (8) Significant at the highest tertile of consumption. This was defined as 106 g/day, but it is not clear if this was the median intake or the lower boundary from the range.
• Balder HF (11) Significant at ≥ 88 g/day.
• Holick CN (18) Significant at 127-188 g/day.

(Non)significant effects at any level of consumption among men were as follows:

• Smith-Warner SA Significant at the 2nd-3rd quintile of consumption, and nonsignificant at the 5th quintile of consumption. No levels of consumption could be defined, because the pooled analysis included study-specific quintiles of consumption from the 6 cohorts.

A clear level of consumption for a protective effect is hard to define, because few RR's differed significantly from 1. But a protective effect may allready be found from consumption of ≥ 127 g/day.

The figure below includes effects from male and/or female cohorts with at least 500 lung cancer cases.

Effect modification by variables other than sex and smoking status:
No evidence of effect modification was found by age of diagnosis, or by the number of fruit and vegetable questions included on a study's FFQ (Smith-Warner SA [24]). And no evidence of effect modification was found by alcohol intake or multivitamin use (Feskanich D [14]).
But in one cohort, the protective effect of fruit was restricted to non-users of beta carotene/vitamin A supplements (Neuhouser ML [22]).

Conclusion: For men & women combined, significant protective effects were found in 6 cohorts, 2 of which were of very large size. These cohorts included a majority of the total amount of cases (61%), and no (non)significantly increased risks were found. The effect was of moderate size (- 15%). Total fruits consumption probably protects against lung cancer risk. Stratified by sex, total fruits consumption probably protects against lung cancer risk among men, while the evidence was only suggestive for a protective effect among women. But since the effect sizes were similar for men and women (- 13, and - 14%, respectively), no heterogeneity in effects seems to exist. The level of consumption for a protective effect is hard to define, but may allready be found at moderate consumption (≥ 127 g/day).

Total fruit and lung cancer risk. Stratified by smoking status.

Background:
About the analysis: For this systematic review, data is stratified by smoking status into current smokers, former smokers, and never smokers.
The JPHC Study (Liu Y [25]) and CARET Study (Neuhouser M [22]) provided data about "ever smokers" without stratifying into former vs current smokers. Likewise, The Finnish Mobile clinic Health Cohort (Knekt P [8]) provided data about "current nonsmokers" without stratifying into nonsmokers vs former smokers. It was chosen not to include this data in the following tables.
About the analysis of current smokers: In 2003 a pooled analysis of 8 prospective studies was published (Smith-Warner SA [24]). The analysis included data from a total of 1,915 smoking cases of which 298 were from The ATBC Study. In 2002 another publication about The ATBC Study included data from 1,644 smoking cases (Holick CN [18]), which is nearly as much as the total amount of cases from the pooled analysis. For the systematic review it was chosen not to include data about current smokers from the pooled analysis as a whole, but from the individual cohorts, because this allowed for a much larger amount of cases in the analysis.

Results:

• Current smokers: Data about current smokers was provided by 8 cohorts, including 6,251 cases. Significant protective effects were found in 3 cohorts of moderate-very large size, including 2,943 cases (47% of all cases). No other (non)significant associations were found. The average RR = 0.86
• Former smokers: Data about former smokers was provided by a pooled analysis of 7 cohorts, and 2 additional cohorts. No (non)significant associations were found, but all RR's were < 1. The average RR = 0.90
• Never smokers: Data about never smokers was provided by a pooled analysis of 7 cohorts, and 3 additional cohorts. No (non)significant associations were found. The average RR = 0.94

Conclusion: Among current smokers, significant protective effects were found in 3 cohorts, including 47% of all cases. Suggestive evidence was found for a protective effect of total fruits against lung cancer risk among current smokers (- 14%). No evidence was found for an association among former or never smokers.

Total fruit and lung cancer mortality.

11 articles, providing information about 9 different cohorts were found. The dietary assessement was conducted following cancer diagnosis, in one cohort only (Skuladottir H [19]).

Results:

• Men: Data was provided about 7 cohorts including 4,057 cases. Significant protective effects were found in 5 cohorts (Wang LD [3], Jansen MC [5], Ozasa K [17], Appleby PN [20], Sauvaget C [21]) including 3,797 cases (94% of all cases). RRs were well below 1 in both remaining cohorts.
  Risk was often presented in frequency of consumption (times or days/week: 3, 9, 17, 20, 21). And significant protective effects were found at (almost) daily consumption in 4 of these cohorts (3, 17, 20, 21). The average RR = 0.60
• Women: Data was provided about 3 cohorts. No (non)significant effects were found, but RRs were < 1.
• Two other cohorts did not stratify risk by sex. No significant effects were found, but RRs were < 1.

Conclusion: Significant protective effects were found in 5 cohorts - one of which of very large size. These 5 cohorts included a total of 84% of all cases. In addition, all RRs were below 1.
Total fruit probably protects against lung cancer mortality. Stratified by sex, (almost) daily consumption of total fruits probably protects against lung cancer mortality among men (- 40%), while no associations were found among women.