Fruits and Vegetables Are Associated with Lower Lung Cancer Risk Only in the Placebo Arm of the β-Carotene and Retinol Efficacy Trial (CARET)¹

Lung cancer is the second most common neoplasm and the leading cause of cancer mortality in both men and women in the United States (1). In 2003, an estimated 170,000 Americans will be diagnosed with lung cancer, and less than 15% of these patients will be alive after 5 years (1). Thus, prevention of lung cancer is an important public health goal (2). The predominant risk factors for lung cancer are cigarette smoking; occupational exposure to radon, asbestos, nickel, chromium, arsenic, bis-chloromethylether, and coal tar; and residential exposure to coal combustion and radon (2, 3). Despite convincing data showing that these exposures, primarily smoking, account for over 90% of all lung cancer cases, there is considerable interindividual variability in disease susceptibility. The reasons for these variations are poorly understood, but there is increasing evidence that numerous endogenous (e.g., carcinogen-metabolizing enzymes) and exogenous (e.g., diet) factors modify the association of the primary risk factors with lung cancer incidence (4, 5). Efforts to understand the complex relationships between established causative agents and factors that may modify lung cancer risk represent an important area of research (6).

Numerous case-control and cohort studies over the last 20 years have investigated associations of diet with lung cancer risk. The majority of the evidence favors an inverse association of fruit and vegetable intake with lung cancer risk, even among smokers (7, 8, 9, 10). For many years, β-carotene was believed to be the principal candidate for the apparent chemopreventive benefit of plant foods because the observational studies showed a consistently strong inverse association for carotenoid-rich foods (e.g., dark green, yellow, and orange vegetables), and investigations using prediagnostic blood specimens found strong inverse associations of serum β-carotene with lung cancer incidence [RR³ = 0.4–0.8 for highest versus lowest quantiles (11, 12, 13)]. Additionally, in vitro, animal, and human experimental data showed that β-carotene could quench free radicals, modulate the immune system, stimulate apoptosis, and enhance cell-cell communication, all of which play a role in carcinogenesis (14, 15, 16, 17). Moreover, extensive evidence from animal and human cell culture studies showed that retinoids were effective chemopreventive agents due to their proliferation-suppressing, differentiation-enhancing influence on epithelial tissues (6). The promising findings from the observational and laboratory studies led to three National Cancer Institute-sponsored chemoprevention trials in Western populations to test whether supplemental β-carotene, retinoids, and, in one study, α-tocopherol would reduce the incidence of lung cancer (18, 19, 20). Remarkably, the Finnish and American trials in high-risk populations showed an increased risk for lung cancer and total mortality attributed to β-carotene (18, 19), and the third, conducted among United States physicians of whom only 11% were current smokers, showed no association between β-carotene supplements and lung cancer risk (20). The unanticipated results of these chemoprevention trials have stimulated considerable scientific interest in learning why β-carotene supplements are carcinogenic rather than chemopreventive agents and whether other dietary factors may reduce lung cancer in high-risk populations. The objective of this report is to examine in detail the association of fruit and vegetable consumption with lung cancer risk in the CARET population.

The CARET was a multicenter randomized, double-blind placebo-controlled chemoprevention trial to test whether daily supplementation with 30 mg of β-carotene and 25,000 IU of retinyl palmitate would reduce the risk of lung cancer among 18,314 heavy smokers and asbestos-exposed workers. The intervention was stopped 21 months early in January 1996, when interim analysis found evidence that the supplements increased the risk of lung cancer and total mortality in this high-risk population by 28% and 17%, respectively (19). Although the CARET intervention ended in 1996, active follow-up of all participants continues until September 2003, including the collection of smoking status and end point data. CARET has excellent retention and follow-up; only 3% of all randomized participants have been lost to follow-up.

Eligible CARET participants were men and women aged 50–69 years who were current or former smokers (within the previous 6 years) with a history of at least 20 pack-years of cigarette smoking (n = 14,254) and men aged 45–69 years (n = 4,060) who were current or former smokers (within the previous 15 years) with occupational exposure to asbestos within the previous 15 years. Evidence for asbestos exposure was either a chest radiograph positive for asbestos-related disease or a work history in specific high-risk trades for at least 5 years. CARET participants were recruited from health insurance rolls, managed care organizations, labor unions, workmen’s compensation programs, and occupational physicians. Participants attended yearly clinic visits and completed follow-up telephone calls every 4 months. Other details about the design and primary results of CARET have been published previously (19, 21, 22). The Institutional Review Board of the Fred Hutchinson Cancer Research Center and each of the five other participating institutions approved all procedures for the study, and participants provided written informed consent at recruitment and throughout the study.

Lung cancer is the primary end point in CARET; detailed outcome data are collected on each case. For each lung cancer case reported to the Coordinating Center at the Hutchinson Cancer Center, clinical records and pathology reports are obtained from the diagnosing physician/hospital for review, and confirmation of tumor origin, location, and histology are determined by the physicians who comprise the CARET Endpoints Review Committee. Lung cancer cases are coded according to the ICDO: adenocarcinoma (ICDO 8140, 8260, 8323, 8480, 8481, and 8550); large cell (ICDO 8012); small cell (8041, 8042, 8044); squamous (ICDO 8070, 8071, and 8073); non-small cell (ICDO 8010); bronchiolo-alveolar (ICDO 8250); and other [ICDO 8000, 8001, 8123, 8240, 8246, 8560, 8801, 8940, and 8980 (23)]. The following participants were excluded from these analyses: 1,845 participants (10%) who enrolled in CARET during the 1985–1988 pilot phase (21) were excluded because of differences in data collection instruments; 1,532 participants (8%) who did not complete a baseline FFQ were excluded; and 817 participants (4%) with implausibly high or low energy intakes (<800 or >5000 kcal/day for men and <600 or >4000 kcal/day for women) were excluded (24). Of the remaining 14,120 CARET participants used in these analyses, there were 742 cases of primary lung cancer after a mean of 8 years of follow-up between 1989 and 2001; 95.8% of these cases are confirmed and closed. Of the open cases, 90% are expected to be confirmed as primary lung carcinomas.

Dietary intake over the previous year was assessed at baseline and every 2 years with a self-administered FFQ, which was reviewed for completion by CARET study staff. For these analyses, we used only the baseline FFQs instead of averaging estimates of intake across several dietary assessments because cases will often have fewer observations than noncases, and the early-diagnosed cases will have fewer observations than the later-diagnosed cases (and hence, lower precision estimates). Thus, to ensure that all participants have the major dietary exposures measured with similar precision, data from the baseline FFQs were chosen for the dietary exposure measurements. To confirm the validity of this approach, we compared fruit and vegetable servings and nutrient estimates from baseline FFQs and FFQs collected 2 years later in a sample of 262 participants. For the dietary exposures examined in this study, mean estimates of intake at baseline were within 1–4% of estimates 2 years later, suggesting a pattern of stable dietary intake over time.

The CARET FFQ was designed to be especially sensitive to the measurement of fruits and vegetables and their nutrients, such as carotenoids. This FFQ is divided into three sections: (a) 7 adjustment questions on types of foods and preparation techniques (e.g., usual types of fat used in cooking and at the table), which are used to alter how the analysis software calculates the fat and fiber content of specific foods; (b) 110 line items with questions on the frequency of use over the last year (from “never or less than once a month” to “2+ per day” for foods and “6+ per day” for beverages) and portion size (small, medium, or large, compared with the stated medium portion size); and (c) 2 summary questions on the usual consumption of fruits and vegetables, which are necessary to reduce the measurement error biased toward overreporting food intake. The nutrient database is derived from the University of Minnesota NCC database (25) and included the 1999 USDA-NCC carotenoid database for United States foods (26).

Plant foods contain thousands of biologically active phytochemicals, such as isothiocyanates, flavonoids, and allyl sulfides, which may protect against cancer (27, 28). Many of these compounds inhibit phase I carcinogen-metabolizing enzymes, induce phase II detoxification enzymes, enhance the immune system, and modulate circulating hormone concentrations (27). Whereas analytic methods exist to quantify many of these phytochemicals (28, 29), few of these data are available in current dietary databases (30). Therefore, to enhance our understanding of the relationships between chemopreventive phytochemicals and lung cancer risk, we created a botanical classification system. Specifically, we grouped the 14 fruits, 19 vegetables, and 12 mixed foods made with vegetables (e.g., pizza, stew) that were listed on the FFQ into seven distinct botanical families plus two “other” groups (see Table 2). For example, frequencies adjusted for portion sizes of the line items “broccoli,” “cauliflower or Brussels sprouts,” “cole slaw, cabbage, sauerkraut,” and “mustard greens, turnip greens, collards” were placed in the “cruciferae” group; and “apples, applesauce, pears,” “peaches, apricots, nectarines,” “strawberries,” and “other juices” comprised the “rosaceae” group. We assigned fractions of servings to mixed foods; for example, “chili with beans” was assigned a half portion of “leguminosae” because kidney beans account for about half of the weight of medium serving of chili. Carrot cake was assigned a one-quarter serving of carrots. Miscellaneous fruits and vegetables that could not be classified into one of the botanical families or for which there was only one line item from that family were grouped into “other fruits” (banana, other berries, papaya, mango) and “other vegetables” (corn, onions, sweet potatoes).

Detailed data on smoking behavior were collected at every CARET contact. This information consisted of current smoking status and smoking history, including age at smoking initiation, total years smoked, and average number of cigarettes smoked/day. For these analyses, we classified participants as either current or former smokers; we estimated total pack-years by multiplying the packs of cigarettes smoked/day times the total number of years smoked at baseline.

Participants completed questionnaires on occupational history. We included a list of eight high-risk trades: asbestos worker; shipyard boilermaker; plasterboard worker; shipyard electrician; plumber/pipefitter; shipscaler; shipfitter; and sheetmetal worker (31). The number of years employed in each high-risk trade and whether the participant was aware of occupational asbestos exposure were recorded.

Age, sex, race/ethnicity, education, general health history, and height and weight were obtained at each participant’s first CARET clinic visit. Body mass index was calculated as weight/height (kg/m²). Dietary supplement use was obtained at each contact. Participants were instructed to take no more than 5500 IU of vitamin A per day and to refrain from any use of β-carotene supplements throughout the study.

Cox proportional hazards models were used to estimate the RRs and 95% CIs for lung cancer. Results are based on an average of 8 years of follow-up of 14,120 CARET participants with time to lung cancer diagnosis as the dependent variable. Multivariate models, based on quintiles of dietary intake among the noncases, were constructed for total fruits and vegetables, total vegetables, total fruit, each botanical group, vitamin C, folate, and the carotenoids. Tests for linear trend across the quintiles were based on a two-sided likelihood ratio test. All models were adjusted for age (in increments of 5 years), sex, race/ethnicity, smoking status, total pack-years of smoking, asbestos exposure, and study enrollment center. RRs associated with specific fruit or vegetable botanical families were adjusted for total fruits or total vegetables using the simple summation method (32). This approach allows us to investigate the effect on lung cancer risk of consuming specific fruits or vegetables (i.e., cruciferae) while keeping total fruit and/or vegetable intake constant. Models including energy adjustment were virtually identical to those without energy adjustment; therefore, to present the most parsimonious model, the final analyses did not include energy adjustment. Other variables that were examined but were neither statistically significant nor influential on RR estimates were body mass index, alcohol consumption, and use of dietary supplements. We used the likelihood ratio test to examine whether the association of dietary intake of fruits and vegetables with lung cancer risk was modified by treatment arm, sex, or smoking status (current versus former smokers). The Ps for the treatment arm by dietary intake interaction tests were modest (P = 0.03–0.07) but fell within the range that warrants the use of a stratified analysis. Therefore, all results are stratified by CARET treatment arm. None of the diet by sex or diet by smoking status interactions was statistically significant (P for interaction values ranged from 0.43 to 0.97), and thus the interaction terms were omitted from the final models. However, as noted above, sex and smoking are included in all models as potential confounding variables. We conducted analyses eliminating lung cancer cases diagnosed within the first year after enrollment in CARET (n = 62) because subclinical disease may affect nutrient intake. These results showed no differences in the estimates of effect; therefore, the final analyses include all lung cancer cases. Subgroup analyses were conducted for histological types of lung cancer to examine whether dietary patterns had different associations with various tumor types.

Table 1 gives the demographic, health, and lifestyle characteristics of the CARET cohort. Lung cancer cases were slightly older, less educated, and more likely to be male compared with noncases. All CARET participants have some smoking exposure, but lung cancer cases were more likely to be current smokers than former smokers and to have more pack-years of smoking compared with noncases. Body mass index and history of asbestos exposure were similar in both groups. There were no differences between the intervention and placebo groups with regard to any of these demographic and lifestyle characteristics (data not shown).

Table 2 gives associations of fruit and vegetable intakes with lung cancer risk. There were 326 and 414 confirmed cases of lung cancer in the placebo and intervention arms, respectively. There were notable reductions in the RR for lung cancer among participants in the placebo arm associated with higher intakes of total fruits, rosaceae fruits, cruciferae vegetables, and “other” vegetables (corn, onions, sweet potatoes). For example, placebo participants who consumed more than 11 servings of fruit per week had a 44% lower lung cancer risk than those who ate less than 2 servings per week (P for trend = 0.003). In contrast, there was no statistically significant decrease in risk among high fruit consumers in the CARET intervention arm. There was a 37% lower risk associated with the highest compared with the lowest quintile of rosaceae fruit consumption in the placebo group (P for trend = 0.02), but there were null findings in the intervention arm. There was a 37% lower risk (P for trend = 0.01) for high compared with low consumers of rutaceae (i.e., citrus) fruits in the placebo arm only (data not shown), but these findings were no longer statistically significant after controlling for total fruit consumption (RR = 0.72; P for trend = 0.15).

Compared with placebo participants who ate one-half serving or less of cruciferae vegetables per week, those who ate three and one-half or more servings per week had a 32% lower risk of lung cancer. Whereas this trend was statistically significant (P for trend = 0.01), the point estimates for each quintile all contain the null value of 1.0. We also note that the point estimate for quintiles two and five are 1.36 and 0.68, respectively, which may partly explain the statistically significant linear trend. Compared with those who ate less than 1.7 servings per week of “other” vegetables (corn, sweet potatoes, onions), placebo participants who ate more than 7.1 servings per week had a 44% reduced risk of lung cancer. Results were null in the intervention arm for both of these vegetable groups. Because there has been considerable interest in the scientific community about potential cancer preventive compounds found in tomatoes and tomato products (33), we examined tomatoes, tomato juice, ketchup, and mixed foods made with tomatoes and tomato sauce separately from the remaining foods in the solanaceae family (see Table 2 footnotes), but the results still showed no association with lung cancer risk in each arm. In models where fruits and vegetables were treated as continuous exposure variables, every increase of 1 serving per week of total fruit, rosaceae fruits, cruciferae, or “other” vegetables was associated with a 3% decrease in risk in the placebo group only (data not shown).

Nutrients that are plentiful in fruits and vegetables (vitamin C, folate, and the carotenoids) were only modestly related to lung cancer risk (Table 3). There was an inverse association of vitamin C with lung cancer risk in the placebo arm when comparing the top quintile with the bottom quintile of consumption; the second and fourth quintiles were nearly identical, possibly due to a threshold effect. There was a suggestion of a similar association for vitamin C in the intervention arm; whereas the trend was statistically significant (P for trend = 0.04), the effect size was smaller than the placebo group, and the 95% CIs for each point estimate include 1.0. The significant trend may be an artifact due to the point estimate of 1.30 for quintile 2 and 0.80 for quintile 5. The carotenoid β-cryptoxanthin, which is found primarily in citrus (rutaceae) fruits, mangoes, peaches, and red peppers, was associated with a 31% lower risk of lung cancer in the placebo arm (P for trend = 0.05); again, findings were null in the intervention arm. Dietary β-carotene, the compound under investigation in CARET, had no association with lung cancer risk in either the intervention or placebo group. We did not observe any statistically significant associations of folate or any of the other carotenoids with lung cancer risk in the intervention arm; there was a near statistically significant inverse association with lycopene (P for trend = 0.08).

We conducted analyses to investigate whether the dietary influences on lung cancer risk varied by tumor type. The majority of cases were non-small cell lung cancers (n = 214 in the placebo arm and n = 281 in the intervention arm). Consistent with the overall study findings, total fruit and rosaceae fruit were each associated with a statistically significant 40% reduced risk of non-small cell lung cancer when comparing the top and the bottom quintile of consumption in the placebo group. There were no statistically significant protective associations in the intervention arm. Cruciferae vegetables were particularly protective against small cell lung cancers. Compared with participants in the bottom quintile, placebo participants in the top quintile of cruciferae consumption had a RR of small cell lung cancer of 0.29 (95% CI, 0.09–0.96), but for those in the intervention arm, the RR was 0.94 (95% CI, 0.44–2.0; data not shown).

In this large chemoprevention trial of heavy smokers, former heavy smokers, and asbestos-exposed workers, consumption of specific classes of fruits and vegetables was inversely associated with lung cancer risk only among participants who were not randomized to receive the CARET study supplements containing 30 mg of β-carotene and 25,000 IU of retinyl palmitate. Among participants in the placebo arm, there were inverse associations of total fruit, rosaceae fruit, cruciferae vegetables, and “other” vegetables (corn, onions and sweet potatoes) with lung cancer risk, but there was no association of any fruits or vegetables with lung cancer risk in the intervention arm. These results were independent of established causative factors for lung cancer (smoking and occupational exposure to asbestos). We interpret our findings as evidence that in this cohort of individuals who are at high risk of lung cancer, plant foods may have an important preventive influence.

Cruciferae vegetables contain dithiolthiones, glucosinolates, indoles, and sulforaphane, which have multiple chemopreventive properties, including inhibition of phase I carcinogen-activating enzymes and induction of phase II carcinogen detoxification enzymes (28, 34, 35). The results presented here showed that compared with CARET placebo participants in the lowest quintile of cruciferae consumption, those in the highest quintile had a RR for lung cancer of 0.68 (P for trend = 0.01). These findings are consistent with data from three case-control studies (4, 36, 37) and one cohort study (38), which found similar magnitudes of effect. We note that, unlike those observational studies, some of which included nonsmokers and light smokers, the CARET population included only heavy current and former smokers and asbestos-exposed workers. Thus, the results from this report add to the evidence that high consumption of cruciferae vegetables may protect against lung cancer, even among individuals who are at substantial risk of the disease.

We know of no studies that have specifically examined rosaceae fruit consumption and lung cancer risk. However, many of the fruits in this botanical family (e.g., apples), as well as onions from the “other” vegetable group, are rich in flavonoids, phytochemicals with strong antioxidant and antitumorigenic properties (27, 39). For example, in vitro studies show that flavonoids promote cell cycle arrest at G₂-M phase and stimulate apoptosis in human cancer cell lines (40, 41). Moreover, quercetin and naringen, two of the most abundant flavonoids in the food supply, inhibit phase I carcinogen-metabolizing enzymes (42, 43) and up-regulate phase II detoxification enzymes (44). Whereas few studies have examined associations of flavonoid intake with lung cancer risk, data exist that support our finding of a multivariate adjusted 37–44% lower lung cancer risk among placebo participants in highest versus lowest quintiles of rosaceae fruit consumption. For example, the Finnish Mobile Health Survey reported an adjusted RR for lung cancer of 0.53 (95% CI, 0.29–0.97) for the highest versus the lowest quartile of flavonoid intake (45). Le Marchand et al. (46) examined specific fruits and vegetables as sources of flavonoids with lung cancer risk and found that consumption of these foods was associated with a 40–50% reduced risk of lung cancer. Taken together, the laboratory, observational, and CARET data support a potentially important role for rosaceae fruits and/or flavonoid-rich foods in lung cancer prevention.

In contrast to previous epidemiological studies (8, 9, 47), including those that motivated CARET (19, 48) and the other β-carotene supplementation trials (18, 20), we did not find an association of baseline dietary β-carotene intake with lung cancer risk. One possible explanation for this discrepancy is the quality of carotenoid data found in food composition databases over the last 20 years. In the 1980s to early 1990s, food carotenoid values were estimated using methods of the American Association of Analytical Chemists (49). These methods were not able to separate individual carotenoids and overestimated pro-vitamin A activity of plant foods, which likely biased previously reported RR estimates for lung cancer. In 1993, a new carotenoid database was published that included values for five carotenoids from 120 foods mostly analyzed by high-performance liquid chromatography (50). Although this database was a substantial improvement, there were no data on mixed foods with fruits and vegetables (e.g., pizza, stew) and no data on carotenoids found in foods such as butter and cheddar cheese, and data for some of the specific carotenoids were very limited or imputed. The 1999 USDA-NCC carotenoid database for United States foods is the most comprehensive and accurate database for carotenoids; it includes a greater variety of foods, including mixed foods, than the 1993 database; uses only values generated by high-performance liquid chromatography; and provides information on different forms of a food (e.g., cooked, raw), which can greatly affect carotenoid content (26). To our knowledge, this is the first report to examine the association of carotenoid intake with lung cancer risk using the 1999 USDA-NCC carotenoid database for United States foods (26). We note that, although we did find a protective association for β-cryptoxanthin, it is possible that this carotenoid is a marker for total fruit, rosaceae, or rutaceae fruit consumption, or it may have independent anticarcinogenic effects (51, 52). Despite the fact that numerous investigations have shown inverse associations of carotenoid intake with lung cancer risk, it is important to remember that there are multiple protective compounds in fruits (e.g., vitamin C, folate, carotenoids) and vegetables (e.g., isothiocyanates, allyl sulfides) and that the mechanisms by which plant foods may decrease cancer risk are likely the cumulative effect of numerous phytochemicals on multiple biochemical pathways.

A particularly interesting finding from this study is that the protective association of fruits and vegetables with lung cancer risk was negated by use of the CARET study supplements containing β-carotene and retinyl palmitate. We propose three possible reasons for the divergent results with regard to fruit and vegetable intake and lung cancer risk from the two study arms. First, experimental studies in animal models show that high doses of supplemental β-carotene modulate the activity of phase I xenobiotic metabolizing enzymes. One study reported a 2–15-fold up-regulation of CYP1A1 and CYP3A1 activity in the lungs of rats given 500 mg of β-carotene per kilogram of body weight, compared with rats given placebo (53). CYP1A1 and CYP3A1 are phase I enzymes that convert numerous carcinogens from cigarettes, including polycyclic aromatic hydrocarbons, into highly reactive electrophiles, which facilitates their conjugation with water-soluble molecules in preparation for elimination from the body. In normal metabolism, these intermediate compounds, which are often more cytotoxic than the parent compounds, are detoxified and eliminated by phase II enzymes that are up-regulated by cruciferae vegetables and flavonoid-rich fruits and vegetables (42, 43, 44). If phase I enzyme activity and subsequent bioactivated carcinogen concentration were increased severalfold by the CARET supplements, then it is conceivable that an excessively large substrate load saturated the detoxification enzyme systems, thus allowing accumulation of carcinogenic intermediate compounds, even at high levels of consumption of foods that up-regulate the phase II enzymes. This scenario could explain the null findings in the CARET intervention arm even for high intakes of cruciferae vegetables and rosaceae fruits.

A second reason for the differential associations of diet with lung cancer risk from the two CARET study arms focuses on factors related to epithelial cell proliferation. Animal model studies with the ferret, whose absorption of β-carotene is more similar to humans than that of the rat, have shown that high-dose β-carotene supplements, particularly in combination with cigarette smoke exposure, down-regulate the RARβ tumor suppressor gene and up-regulate c-Fos, c-Jun, and cyclin D1, which induce cell proliferation (54, 55, 56). The net effect in these animal model systems has been a strong proliferative response and evidence of lung carcinogenesis (55). Notably, whereas high-dose β-carotene may up-regulate expression of these cell cycle-regulatory proteins, flavonoids are inhibitory (57). In vitro (40), animal model (57), and human feeding studies (41) all support an important role for flavonoids in signal transduction and cell cycle regulation, in addition to their previously noted antioxidant and phase I and II enzyme-regulatory functions. The CARET supplements could have promoted up-regulation of c-Fos, c-Jun, and cyclin D1 such that even high levels of flavonoid-rich fruits and vegetables (e.g., onions, rosaceae fruit) were unable to override the strong proliferative signals. Furthermore, Salgo et al. (58) reported that β-carotene, including its epoxide and ketone derivatives, increased benzo(a)pyrene adduct formation in vitro.

A third, but least likely, reason for the discrepant findings between the study arms presented in this report relates to whether the β-carotene in the CARET supplements adversely affected the bioavailability of other dietary carotenoids. Whereas this may be an attractive hypothesis because carotenoids compete for incorporation into micelles and subsequent absorption, there is little evidence to support it. Previously published findings from both CARET and the α-Tocopherol β-Carotene Cancer Prevention Study showed no adverse effects of β-carotene on serum concentrations of the other carotenoids and no excess of either lung cancer incidence or total mortality among participants in the highest quintile of baseline serum β-carotene concentration (48, 59, 60). Additional studies to examine the effects of dietary phytochemicals, pharmacological doses of micronutrients, such as β-carotene, and known causative agents for lung cancer on cell cycle regulation may provide a greater understanding of these complex biological mechanisms and their relation to disease risk.

Few other studies have had sufficient numbers of lung cancer cases to stratify by tumor histology; two of three investigations showed that dietary factors may have different associations across histological cell types (61, 62, 63). Our results showing that total fruit, rosaceae fruits, and cruciferous vegetables were strongly protective against non-small cell and small cell lung cancers provide additional data to suggest that dietary factors may have discriminating protective capabilities for various tumor types. These data should be considered preliminary but may provide interesting hypothesis-generating opportunities about the relationships of phytochemicals with tumor characteristics. Large cohorts with sufficient numbers of subtypes of lung cancer are needed to test such hypotheses. In the future, it is likely that molecular classifications of lung cancers (and other cancers) will enhance or eventually supersede histological classifications (64, 65, 66, 67).

This study has several strengths. CARET is a large randomized intervention trial that carefully collected detailed smoking, dietary, and outcome data. Thus, we had sufficient power to detect moderate associations of the dietary exposures with lung cancer risk, and the likelihood of residual confounding from unmeasured smoking behavior or other risk factors for lung cancer is small. All reports of lung cancers are confirmed by medical records and pathology reports, which are carefully reviewed by the CARET Endpoints Review Committee. There are also limitations that must be mentioned. First, because most lung cancers are not diagnosed until the disease is advanced, there may be undetected lung carcinomas present in individuals who were classified as noncases. Second, our discussion of the results presented in this report has focused on β-carotene, primarily due to the totality of evidence from CARET and other intervention trials (6). However, because the CARET study supplements were a combination of β-carotene and vitamin A, we are unable to distinguish any adverse effects from the two compounds. Another limitation is that the nutrient intake data were estimated by self-report from a FFQ, which is subject to both random and systematic bias (24). There was no formal validation of the CARET FFQ with other dietary assessment instruments, such as food records or recalls. Finally, our results cannot be generalized to the entire population but are most applicable to those at high risk of lung cancer.

The principal causative agents of lung cancer are exposure to cigarette smoke, asbestos, radon, and several industrial chemicals. Avoidance of these substances could prevent almost all cases of lung cancer and should remain the primary focus of prevention efforts. Nonetheless, the findings from this report show that, among heavy smokers and asbestos-exposed workers, consumption of fruits and vegetables, especially from the rosaceae and cruciferae families, and vegetables, such as onions and corn, may lower risk by 32–44%, a magnitude of reduction that has substantial public health implications. However, this report shows that persons at risk for lung cancer who use β-carotene supplements do not benefit from the compounds in plant foods, probably due to interference of the supplements with the effects of bioactive phytochemicals. These results may provide a partial explanation for the adverse and no effect lung cancer incidence findings reported from randomized clinical prevention trials of β-carotene supplementation in CARET, α-Tocopherol β-Carotene Cancer Prevention Study, and the Physicians’ Heath Study, respectively. Finally, this report, the first to use the 1999 USDA-NCC carotenoid database for United States foods in relation to lung cancer, finds no basis for the widely accepted view that there is an inverse association of dietary β-carotene with lung cancer risk.

We thank Dr. Marjorie Perloff (National Cancer Institute) and the CARET participants, staff, and investigators for important contributions to this work.