INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Bronchogenic carcinoma is currently the leading cause of cancer death in the United States. In 1998 there will be an estimated 160,100 deaths attributable to lung cancer (1). In the United States, the two major known risk factors for bronchogenic carcinoma are smoking and asbestos exposure (2, 3). The tumors that arise in association with cigarette smoking tend to be in the upper lobes, with a typical upper to lower lobe ratio of roughly 2.5:1.0 (4). For example, among the 15,477 histologically proven cases of primary lung cancer among white men in the Surveillance, Epidemiology, and End Results (SEER) Study, 65% of the tumors originated in the upper lobes compared with 35% in the middle or the lower lobes (5). The data are more conflicting regarding the association between asbestos exposure and lung cancer location. Among asbestos-exposed individuals, some studies have reported an inversion of this ratio, with a higher proportion of tumors occurring in the lower lobes (7), but others have reported a predominance of upper lobe tumors, a pattern similar to lung cancer arising in the general population with cigarette exposure (6, 11). Since lung cancers associated with smoking tend to occur in the upper lobes and smoking and asbestos exposure have multiplicative effects on the risk of lung cancer (14, 15), it is not surprising that asbestos exposure would also be associated with an upper lobe location of lung cancer. In addition, the upper lobe predominance of lung cancers associated with asbestos exposure is consistent with reports that the asbestos fiber concentration may be higher in the upper lobes than in the lower lobes (16).

The pathophysiologic basis for the predominance of upper lobe lung cancers among smokers (and likely also among asbestos-exposed individuals) is not well understood, but a similar upper lobe predominance is observed for emphysema among smokers. For both emphysema and lung cancer, an upper lobe predominance is observed despite the fact that a higher proportion of the ventilation occurs in the lower lobes. It is possible that toxins and carcinogens may persist longer in the upper lobes due to less relative ventilation or less efficient lymphatic clearance.

Alternatively, upper lobe predominance of lung cancer may be, in part, due to less efficient delivery of diet-derived or other protective substances via the circulation to the upper lobes compared with the lower lobes. Numerous cohort and case-control studies have consistently reported an inverse relationship between intake of certain vegetables and fruits and lung cancer risk (17) as well as a significant inverse relationship between serum -carotene levels and lung cancer risk (20). Similar relationships have been reported with intake of vitamins C (23) and E (21, 22, 26). These observations lend support to the hypothesis that a diet-derived circulating factor with possible antioxidant properties may be protective against lung cancer. Although recent prospective, randomized, controlled trials did not find supplemental intake of -carotene to be protective against development of lung cancer in high-risk individuals (27, 28), it is still likely that other substances in vegetables and fruits are beneficial. Because the ventilation-perfusion (/) ratio is higher in the upper lobes compared with the lower lobes, we hypothesized that the balance between cigarette-derived carcinogenic substances delivered via the airways and diet-derived protective substances delivered via the circulation would be less favorable in the upper lobes compared with the lower lobes. If so, individuals who consume less of certain vegetables, fruits, and vitamins may not only be expected to have an increased risk of lung cancer, but also an increased risk of upper lobe tumors, in particular. In order to test our hypothesis, we examined the data from a large case-control study of 328 lung cancer patients in whom detailed dietary, occupational, and smoking histories were available.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

This study is part of a large case-control study of incident cases of lung cancer at the Massachusetts General Hospital (MGH), which involved genotyping for metabolic polymorphisms and determination of DNA adduct levels from peripheral leukocytes and lung tissue (29). Eligible cases included all patients with newly diagnosed primary lung cancers presenting to the thoracic surgery service at MGH between December 1, 1992 and August 15, 1996. All cases were confirmed by histologic examination. Almost all of the case subjects had surgically resectable tumors (stage I, II, or IIIA); fewer than 2.7% had stage 3B or stage 4 cancers. The study was approved by the Human Subjects Committees at MGH and the Harvard School of Public Health. Of the 562 eligible cases during the study period, 42 refused to participate, 34 were missed, 26 were unable to participate, and four agreed to participate but did not complete the questionnaires. Of the remaining 456 (81%) cases enrolled in our study, 328 (72%) had complete dietary, occupational, and smoking histories and were included in this analysis.

Demographic information from cases and controls (gender, age, race, education, family history of cancer) was collected by an extensive questionnaire administered by trained interviewers at the hospital. The race of the subjects was categorized as either white or nonwhite. Subjects with at least some college education were considered to have a high level of formal education and those completing up to high school education were considered to have a low level of formal education. A subject with any biological parent or sibling with a history of any known cancer, other than nonmelanomatous skin cancers, was considered to have a family history of cancer. A subject with any biological parent or sibling with a history of lung cancer was considered to have a family history of lung cancer.

Smoking history and other occupational or environmental exposure information were collected using a modified standardized American Thoracic Society respiratory questionnaire (30). This questionnaire included information on current smoking status, total pack-years of smoking exposure, age at which smoking started, number of years since quitting smoking, and period and frequency of exposure to other environmental or occupational substances, such as asbestos. Subjects were assigned a current smoking status based on whether they had never smoked (nonsmokers), had not smoked for more than 1 yr (ex-smokers), or were smoking at the time of the study (current smokers). Those who had quit smoking for less than 1 yr were classified as current smokers for the purposes of this analysis. The number of years since quitting smoking was considered to be zero for current smokers, and the age of the subject was used for this value for those who had never smoked. Subjects were divided evenly into two groups according to pack-years of smoking exposure: those in the lower stratum had less than 52.2 pack-years and those in the higher stratum had 52.2 or more pack-years. Primary tumors located in the right lower lobe, right middle lobe, or the left lower lobe were classified as lower lobe tumors. Primary tumors located in the right upper lobe or the left upper lobe were classified as upper lobe tumors. Patients with multiple tumors were excluded unless all of the tumors could be localized either to the lower or the upper lobes. Tumors were classified as squamous cell, large cell, small cell, or adenocarcinoma. Patients with tumors of mixed histology were excluded from the analysis.

Asbestos exposure was assessed using a previously described asbestos exposure index (31). This index was derived based on knowledge of asbestos exposure in the New England building construction trades. The heaviest asbestos exposure occurred before 1965. By 1965 the use of fiberglass insulation for new insulation application became well established and widespread. Therefore, after 1965 exposure to asbestos occurred mainly during repair, remodeling, or renovation work. After 1972 the use of asbestos for new insulation and fireproofing was discontinued. The promulgation of the permissible exposure limit for asbestos by the Occupational Safety and Health Administration in 1972 also helped to further reduce workplace exposures. Thus, the weighted duration of asbestos exposure was calculated for each subject based on the duration of work during these three time periods: a weight of 4 was given to each year of asbestos exposure prior to 1965; a weight of 2 was given to each year worked from 1965 to 1972; and a weight of 1 was given to each year worked after 1972. In addition, specific jobs were assigned a weight (intensity factor) ranging from 4-6 depending on the type of exposure. For example, a weight of 4 was assigned to the following types of exposure: automobile repair, brake mechanic, building maintenance, carpentry, demolition of buildings, drywall hanging, fire fighting, smelting, tunnel construction, and welding; a weight of 5 was assigned to the following types of exposure: boiler-making, foundry work, iron/steel manufacturing, pipe fitting, and construction work; a weight of 6 was assigned to the following types of exposure: insulation installation, pipe covering/insulating, ship building/repair. Finally, a cumulative index (asbestos exposure index) representing the intensity as well as the weighted duration of asbestos exposure was calculated for each subject by multiplying the number of years of exposure, the weight based on specific years of exposure, and the intensity factor. Thus, a subject who had worked year-round for 2 yr as a shipbuilder before 1965 would have an index score of 48 (2 × 4 × 6), and a subject who had worked year-round as an automobile mechanic for 3 yr after 1972 would have an index score of 12 (3 × 1 × 4). Based on this score, subjects with a score of more than 20 were defined to have had significant exposure and those with a score of 20 or lower were defined to have had nonsignificant exposure to asbestos. The cut-off point was selected based on our observation of a step-up increase in lung cancer risk when the asbestos exposure score exceeded 20 and the lack of a demonstrable increase in lung cancer risk when the score was 20 or lower.

Dietary information was estimated using a semi-quantitative food frequency questionnaire that included information about 126 food items (32). Subjects were asked how often they consumed each food item in the year before lung cancer diagnosis. For each food item, a commonly used unit or portion size was specified as a typical serving, such as one-half cup of carrots or one small glass of orange juice. There were nine possible responses: (1) never or less than once a month; (2) 1-3 times per month; (3) once per week; (4) 2-4 times per week; (5) 5-6 times per week; (6) once per day; (7) 2-3 times per day; (8) 4-5 times per day; (9) 6 or more times per day. Vegetable food items were grouped as yellow-orange vegetables or dark green vegetables. A serving of yellow-orange vegetables was defined as one-half cup of raw or cooked carrots, yellow squash, zucchini, yams, or chili sauce. A serving of dark green vegetables was defined as one-half cup of broccoli, brussels sprouts, spinach, kale, mustard, or chard greens. A serving of fruit was defined as one-half cup of apples, pears, bananas, cantaloupe, oranges, grapefruit, strawberries, blueberries, peaches, apricots, or plums. Intake of vitamins A, C, and E were estimated using data from the current United States Department of Agriculture sources (33) and other published information. Intake of carotenoids and animal fat were computed using the food composition data from the Nutrient Composition Laboratory of the United States Department of Agriculture (33). Subjects were divided evenly into tertiles according to their level of yellow- orange vegetable consumption: those in the lowest tertile had intake of less than 0.30 servings per day; those in the middle tertile had intake of 0.30 or more but less than 0.58 servings per day; and those in the highest tertile had intake of 0.58 or more servings per day. Subjects were similarly divided into tertiles according to their level of vitamin E intake.

Initially, univariate analysis was used to examine the potential associations of age, height, weight, sex, race, family history of cancer, family history of lung cancer, level of education, tumor histology, smoking history, asbestos exposure history, and dietary history with the location of the tumor. The two-tailed t test for unpaired data, Wilcoxon rank sum test, Fisher exact test, and Mantel-Haenszel test for trend were used to compare the means of continuous normal variables, continuous non-normal variables, categorical variables, and ordinal variables, respectively. The associations between yellow-orange vegetable consumption, vitamin E intake, and tumor location were examined after stratifying patients according to pack-years of smoking exposure.

The variables found to have significant or near significant (p  0.10) association with tumor location in univariate analysis were included in a multivariable logistic regression model. All statistical analyses were performed on SAS software (SAS Institute Inc., Carey, NC).

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The descriptive characteristics of the patients included in the analysis are summarized in Table 1. There were 235 patients with upper lobe tumors (71.6%) and 93 patients with lower lobe tumors (28.4%), giving an overall upper to lower lobe ratio of 2.5:1.0. There were no significant differences between the two groups in regard to age, height, weight, sex, race, education level, family history of cancer, or tumor histology. However, family history of lung cancer, smoking history, and asbestos exposure history were significantly different between those with upper versus lower lobe cancers. Patients with upper lobe tumors were more likely to have a family history of lung cancer than those with lower lobe tumors (24.8% versus 12.5%, p = 0.02). Patients with upper lobe tumors also had more exposure to tobacco based on mean pack-years smoked (58.6 versus 49.9, p < 0.03) and less mean time since quitting smoking (8.5 versus 10.4 years, p = 0.10). In addition, a higher proportion of patients with upper lobe tumors had a history of significant exposure to asbestos (14.5% versus 4.4%, p = 0.01).

The association between tumor location and diet is summarized in Table 2. Daily mean consumption of total calories and animal fat did not differ significantly between the two groups. However, the mean intake of yellow-orange vegetables (0.49 versus 0.68 servings per day, p = 0.002), -carotene (0.72 versus 0.96 mg/d, p = 0.004), -carotene (5.56 versus 6.96 mg/d, p = 0.003), and vitamins A (14410 versus 18117 IU/d, p = 0.004), C (288 versus 386 mg/d, p = 0.01), and E (63 versus 124 mg/d, p < 0.05) were significantly lower in patients with upper lobe tumors. The intake of dark green vegetables (0.34 versus 0.41 servings per day, p = 0.09), and fruits (2.79 versus 3.33 servings per day, p = 0.10) also tended to be lower in subjects with upper lobe tumors, but the differences were not statistically significant.

In multivariable logistic regression analysis (see Table 3), the independent predictors of tumor location were history of asbestos exposure (p = 0.02), family history of lung cancer (p = 0.03), yellow-orange vegetable intake (p < 0.04), and vitamin E intake (p = 0.05). Intake of dark green vegetables, fruits, and vitamin C were no longer significant predictors of tumor location in multivariable analysis. Adjustment of the model for total caloric intake did not change the significance of the association between tumor location and history of asbestos exposure (p < 0.04), family history of lung cancer (p = 0.02), yellow-orange vegetable intake (p < 0.04), and vitamin E intake (p < 0.04).

The Spearman correlation coefficients between intake of yellow-orange vegetables and intake of green vegetables, fruits, -carotene, -carotene, vitamin A, vitamin C, and vitamin E were 0.45 (p < 0.001), 0.37 (p < 0.001), 0.76 (p < 0.001), 0.76 (p < 0.001), 0.75 (p < 0.001), 0.36 (p < 0.001), and 0.25 (p < 0.001), respectively. Thus, as expected, the intake of yellow-orange vegetables was not independent from these other dietary variables. Similarly, the Spearman correlation coefficients between vitamin E intake and intakes of -carotene, -carotene, dark green vegetables, fruits, and vitamins A and C were 0.16 (p < 0.01), 0.45 (p < 0.001), 0.19 (p < 0.001), 0.28 (p < 0.001), 0.59 (p < 0.001), and 0.63 (p < 0.001), respectively. Thus, as expected, vitamin E intake was not truly independent from these other dietary variables. However, the associations between tumor location and intakes of yellow- orange vegetable and vitamin E remained significant whether none, some, or all of these other dietary variables were included in the logistic regression analysis (data not shown).

The ratios of upper to lower lobe tumors were 3.6, 3.0, and 1.6 for those in the lowest, middle, and highest tertiles of yellow-orange vegetable intake, respectively (p = 0.008). The crude and smoking-adjusted odds ratios of upper lobe tumor location for patients in the highest versus lowest tertile of yellow-orange vegetable consumption was 0.45 (0.24-0.82, p = 0.01) and 0.47 (0.26-0.88; Cochran-Mantel-Haenszel test, p < 0.02; Breslow-Day test for homogeneity of the odds ratio, p > 0.40), respectively. Figure 1 demonstrates the relationship between yellow-orange vegetable consumption and tumor location, with subjects stratified by smoking exposure. For patients in the lower stratum of smoking exposure, the ratios of upper to lower lobe tumors were 3.2, 2.0, and 1.2 for those in the lowest, middle, and highest tertiles of yellow-orange vegetable intake, respectively (p = 0.02); among those in the higher stratum of smoking exposure, the ratios were 3.9, 4.1, and 2.5, respectively (p > 0.40). Within each tertile of yellow-orange vegetable intake, those in the higher stratum of smoking exposure had a higher ratio of upper to lower lobe tumor ratio compared with those in the lower stratum. Thus, the likelihood of the tumor occurring in the upper lobes tended to be significantly higher for those with a combination of greater smoking exposure and less yellow-orange vegetable intake compared with those with a combination of less smoking exposure and greater yellow-orange vegetable consumption (p = 0.001) (Figure 1).

View larger version (29K): [in this window] [in a new window]   Figure 1.   Relationship between yellow-orange vegetable consumption, smoking, and ratio of upper:lower lobe tumors. Mantel-Haenszel test for trend, p = 0.001.

The ratios of upper to lower lobe tumors were 2.7, 3.9, and 1.7 for those in the lowest, middle, and highest tertiles of vitamin E intake, respectively (p = 0.08). Figure 2 demonstrates the relationship between vitamin E intake and tumor location, with subjects stratified by smoking exposure. For patients in the lower stratum of smoking exposure, the ratios of upper to lower lobe tumors were 2.3, 2.7, and 1.2 for those in the lowest, middle, and highest tertiles of vitamin E intake, respectively (p = 0.08); among those in the higher stratum of smoking exposure, the ratios were 3.1, 5.7, and 2.6, respectively (p > 0.75). Within each tertile of vitamin E intake, those in the higher stratum of smoking exposure had a higher ratio of upper to lower lobe tumor ratio compared with those in the lower stratum. Thus, the likelihood of the tumor occurring in the upper lobes tended to be significantly higher for those with a combination of greater smoking exposure and less vitamin E intake compared with those with a combination of less smoking exposure and greater vitamin E intake (p = 0.003) (Figure 2).

View larger version (27K): [in this window] [in a new window]   Figure 2.   Relationship between vitamin E intake, smoking, and ratio of upper:lower lobe tumors. Mantel-Haenszel test for trend, p = 0.003.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

For patients with predominantly surgically resectable lung cancer (< 2.7% stage IIIB or IV), we found a strong inverse association between upper lobe tumor location and intake of yellow-orange vegetables and vitamin E. Patients with upper lobe tumors consumed fewer yellow-orange vegetables and less vitamin E than those with lower-lobe tumors. In addition, an exposure-response relationship could be demonstrated between the proportion of patients with upper lobe tumors and yellow-orange vegetable consumption. The proportion of upper lobe tumors decreased in a stepwise fashion with increasing yellow-orange vegetable consumption. However, a similar monotonic exposure-response was not observed for vitamin E intake. In multivariable logistic regression analysis, which adjusted for smoking (current smoking status, pack-years of smoking, and time since quitting smoking), asbestos exposure, family history of lung cancer, and other dietary factors (intake of dark green vegetables, fruits, and vitamin C), the inverse relationships between upper lobe tumor location and intakes of yellow-orange vegetables and vitamin E remained significant (p < 0.04 and p = 0.05, respectively). Further adjustment of the model for intake of total calories, -carotene, -carotene, and vitamin A did not change the significance of the associations between tumor location and intake of yellow-orange vegetables and vitamin E. The odds ratios of upper lobe tumor location for those in the highest versus lowest yellow-orange vegetable consumption were similar whether compared using univariate (0.45), stratified (0.47), or multivariable logistic regression analysis (0.50).

None of the other dietary variables, including intake of fruits, dark green vegetables, -carotene, -carotene, total calories, animal fat, or vitamins A and C were independently predictive of tumor location in a multivariable logistic model. The lack of significant association between tumor location and intake of -carotene or vitamin A in our study is consistent with prior randomized, controlled trials, which found no significant benefit of supplemental -carotene or vitamin A intake and reduction of lung cancer risk (27, 28). Alternatively, these other dietary variables may be inversely associated with tumor location but do not add further information in the multivariable model because they are highly correlated with yellow- orange vegetable consumption and vitamin E intake.

Our data show that subjects who consume fewer yellow- orange vegetables or less vitamin E are more likely to develop upper lobe tumors than lower lobe tumors. Furthermore, the likelihood of the tumor occurring in the upper lobes tended to be significantly higher for those with a combination of greater smoking exposure and less yellow-orange vegetable intake compared with those with a combination of less smoking exposure and higher yellow-orange vegetable intake (Figure 1). Similarly, the likelihood of the tumor occurring in the upper lobes tended to be significantly higher for those with a combination of greater smoking exposure and less vitamin E intake compared with those with a combination of less smoking exposure and greater vitamin E intake (Figure 2). The exact mechanism(s) for these relationships cannot be determined from this study. However, these findings are consistent with our hypothesis that the balance between cigarette-derived carcinogens delivered via the airways and diet-derived protective substances delivered via the circulation is less favorable in the upper lobes compared with the lower lobes, just as the / ratio is higher in the upper lobes compared with the lower lobes. If so, subjects who consume fewer vegetables, fruits, or vitamins would not only be expected to have a higher risk of lung cancer, as has been shown in many prior observational studies (17, 34), but a higher risk of upper lobe tumors in particular, as shown in our study.

In stratified analysis, the inverse association between upper lobe tumor location and intake of yellow-orange vegetables was significant for those in the lower stratum of smoking exposure ( 52 pack-years) but not for those in the higher stratum (> 52 pack-years). For those in the lower stratum of smoking exposure, it is possible that the amount of cigarette-derived carcinogens delivered via the airways is modest enough for diet-derived substances delivered via the circulation to have a protective effect, particularly in the lower lobes. On the other hand, for those in the higher stratum of smoking exposure, it is possible that the amount of cigarette-derived carcinogens delivered via the airways is too high for diet-derived substances to have a significant protective effect, even in the lower lobes. If so, it is interesting to speculate whether chemoprevention trials targeting subjects with more moderate smoking exposures would have been more fruitful than the published trials, which have targeted generally higher risk patients (27, 28).

There are several potential limitations to our study. Almost all of our case subjects had operable tumors (stage I, II, or IIIA); fewer than 2.7% of the patients had stage IIIB or IV cancers. However, there is no reported difference in the proportion of operable tumors based on the lobe-of-origin of the tumor (35). Although cancer patients who are malnourished may be poorer surgical candidates, we found no significant differences in patient weight or mean daily calorie intake between the two groups. Therefore, it is unlikely in this population that selection of mainly operable lung cancers would have biased the study in terms of tumor location. Because our case subjects primarily had operable tumors, less than 3% of the patients had small-cell lung cancers. Although there is no reason to expect different ratios of upper to lower lobe tumors based on histology, our findings should be applied primarily to non-small-cell lung cancers. Future studies that include more patients with small-cell lung cancers would be informative.

Another limitation of the study is that dietary history was based on a questionnaire and thus subject to recall bias. However, for such bias to have significantly affected our findings, the bias would have to have affected the two patient groups differentially. There is no a priori reason why tumor location might influence recall. Measurement of serum carotene levels may have helped confirm the accuracy of dietary information to some degree, but serum levels at time of illness may not accurately reflect the dietary patterns in years past, which is likely more relevant for carcinogenesis. Nevertheless, confirmation of our findings in a prospective cohort study where dietary information is gathered in advance of illness would be informative.

Of the 456 cases enrolled in our study, 328 (72%) had complete dietary, occupational, and smoking histories. Thus, limiting our analysis to these 328 cases could have caused some selection bias. The ratio of upper to lower lobe tumors was 3.1:1.0 for those excluded from our analysis because of incomplete dietary, occupational, or smoking histories. This ratio was not statistically different than the ratio of 2.5:1.0 for the 328 cases included in our analysis (p > 0.55). Thus, there is no reason to suspect that exclusion of those with incomplete information would have biased the study.

The pathophysiologic basis for predominance of upper lobe tumors among smokers is not well understood. To our knowledge, no prior study has examined the association between diet and lung cancer location. In our analysis of 328 patients with predominantly surgically resectable lung cancer, there were strong inverse relationships between upper lobe tumor location and intake of yellow-orange vegetables and vitamin E. Thus, our results suggest that the location of lung cancer may be determined, in part, by dietary factors.