Vol 25, No. 1 Printed in Great Britain

International Journal of Epidemiology O International EpHdemlologlcal Association 1996

Dietary Factors and Lung Cancer among Men in West Sweden GOSTA AXELSSON,* TOMI LILJEQVIST,* LARS ANDERSSON," BENGT BERGMAN* AND RAGNAR RYLANDER* Axelsson G (Department of Environmental Medicine, G6teborg University, Medlcinaregatan 16, S-413 90 Gothenburg, Sweden), Uljeqvist T, Andersson L, Bergman B and Rylander R. Dietary factors and lung cancer among men in West Sweden. IntemationalJoumal of Epidemiology 1996; 25: 32-39. Background. Previous studies have reported an association between tea drinking and lung cancer. In view of these data, the relationship between tea drinking as well as other dietary factors and lung cancer was investigated in a case-control study In the west of Sweden. Methods. Patients with suspected lung cancer were collected from pulmonary units at central hospitals in the area investigated, and population controls were matched for age. The material reported here comprises 308 male cases with a confirmed diagnosis of lung cancer and 504 controls. The participants were interviewed by specialty trained nurses, using a questionnaire to assess smoking, dietary habits, occupational exposures and conditions in the residential area (local air pollution). This paper reports the results from dietary factors studied with a food frequency technique. Results. The results demonstrated a strong protective effect of vegetables (odds ratio [OR] = 0.69, 95% confidence interval [Cl]: 0.46-1.05, and OR = 0.37, 95% Cl : 0.23-0.61 for Intermediate and high consumption classes respectively). A low OR was consistent for all hlstologlcal types of lung cancer. High consumption of fruits did not show any similar protective effect. Drinking milk was associated with a dose-response related risk increase after adjustment for smoking and vegetable consumption (P for trend = 0.07). Odds ratio was 1.73, 95% Cl : 1.00-3.01 for high consumption of milk. Conclusions. High intake of vegetables had a strong protective effect among males. Diet is thus a potential confounding factor in studies on lung cancer and environmental factors and should thus be taken into consideration in the planning of such studies. Keywords: lung cancer, diet, vegetables

There is overwhelming epidemiological evidence that certain dietary factors are related to the risk of developing lung cancer. Since the first study in 1975 on vitamin A and lung cancer,1 about 40 studies on dietary factors have been published, of which several have been analysed in two major reviews.2"3 Most of the studies show that fruit and vegetables are protective factors, and some studies suggest that fat4"6 and cholesterol7 increase the risk. The different studies show certain inconsistencies. For some factors, there are conflicting results on protection or increased risk. This is hardly surprising, considering the methodological crudeness of certain studies and that they involve the study of low risk agents. 8 Particularly important methodological shortcomings in these studies are the variations induced by the sources of information (the case itself or next of kin),

the drop out from the selected cases and poor participation among the control groups. Although several studies have been published on dietary factors and lung cancer, differences in dietary habits across different ethnic groups justify further studies in specific populations to delineate local risk factors. Such information is also required for successful public health programmes. Another reason for new studies in a population group is the information impact of local studies, in view of the rather limited general knowledge of the relationship between dietary factors and lung cancer, even among physicians. The rationale for the present study derives from data suggesting tea as a risk factor for lung cancer. Three studies have detected an association between drinking tea and an increased risk of developing lung cancer.9"11 In the latter study,11 experimental investigations demonstrated that tea extracts were mutagenic. In view of these data, a case-control study was undertaken, in the west of Sweden, to investigate the possible risk for lung cancer associated with tea drinking. Other dietary factors which were known or suspected at the time to influence the risk of developing lung cancer

• Department of Environmental Medicine, GQteborg University, Medicinaregatan 16, S-413 90 Gothenburg, Sweden. • • Lung Clinic, North Alvsborg Genera] Hospilal, Trollhattan, Sweden. * Department of Pulmonary Medicine, Sahlgrenska Hospital, Gothenburg, Sweden.

32

DIETARY FACTORS AND LUNG CANCER

were also studied to obtain information on risk and protective factors relevant to the Swedish population. In addition to dietary factors, smoking habits, occupational exposure, residential areas in terms of air pollution and keeping birds were studied as potential risk factors. This report contains the results from the analysis of dietary factors and lung cancer among males. As the number of female lung cancer cases is still too small, a separate report about females will be written only after an additional year of data collection.

MATERIALS AND METHODS Study Population and Case-Control Selection The study base comprises people =575 years of age of Scandinavian birth who were residing in one of 26 municipalities in Gothenburg and Bohus county and Alvsborg county in the southwest of Sweden. The municipalities were selected to represent the area from which patients with suspected lung cancer were referred to the pulmonary units at the regional hospitals. Routines were established to identify suspected lung cancer cases at three hospitals in the region: Renstromska Hospital in Gothenburg and the Departments of Pulmonary Medicine at the North Alvsborg General Hospital in Trollhattan and Boris General Hospital. These hospitals cover more than 90% of the lung cancer cases in the Gothenburg area and 75% of the cases in the county of Alvsborg. Patients who were referred to one of these hospitals and suspected of having lung cancer (on the basis of lung X-ray findings) were invited to participate in the study. A regular survey was also made of inpatients to ensure that lung cancer cases in the study base who had been transferred from other hospitals were included in the material. The patients received written and oral information concerning the project, which was described as a research project aimed at studying the relationship between environmental factors and lung disease. The letter emphasized the patients' voluntary participation and confidentiality concerning the handling of data. Patients who were willing to participate in the project were immediately registered at the Department of Environmental Medicine, after which one of the interviewers contacted the patient to make an appointment for an interview. To select controls, a list of the personal identification numbers (a number determined by birth date) of all suspected lung cancer cases in the study base (patients who were willing to take part as well as those who had declined) was sent, approximately every fourth week, to the regional population registration

33

authority in Gothenburg and Bohus county and Alvsborg county. For each patient, the next person in the respective county who was of the same sex as the patient and closest to the patient in the order of the personal identification number was identified as control. If it was detected through the name or personal identification number that the individual was an immigrant, a second person was selected as the control. Each control person received a letter with the same contents as that presented to the patients with suspected lung cancer. Information was also given as to how controls were chosen. If a selected control became a nonrespondent, a substitute was not selected. The controls received a lottery ticket in the State lottery for their participation. Of the 642 suspected male cases of lung cancer, 568 (88%) were interviewed. Of the 74 patients (12%) who were not interviewed, 35 had accepted participation but were too ill to undergo the interview, four were not asked on the grounds of mental disease and 35 declined. Among the 644 controls, 140 (22%) were not interviewed: 29 declined because of ill health, 55 refused without indicating a reason and 56 did not answer the letter or telephone call. Twice a year, the suspected cases and controls were matched with the regional cancer registry. They were classified as lung cancer cases (ICD-7, 162.1) only if they had a complete notification in the registry. According to the regulation of the National Board of Health and Welfare in Sweden, all physicians in hospitals and other establishments for medical treatment must report all cases of diagnosed cancer to the cancer registry. Furthermore, pathologists and cytologists separately report every cancer diagnosis on surgically removed tissue, biopsies, cytological specimens and autopsies. Thus cases are usually recorded in the registry from two separate sources. Of the 568 male suspected cases interviewed, 54% were later diagnosed as having primary lung cancer. Those patients who did not receive this diagnosis were excluded from the analysis and not used as controls. Of those who were too ill to participate or who declined, 49% were later diagnosed as having primary lung cancer. Three men who were interviewed as population controls were later reclassified as cases, as they received a diagnosis of lung cancer about one year after the interview. Questionnaire The questionnaire used at the interview included questions on smoking, environmental tobacco smoke (ETS),

34

INTERNATIONAL JOURNAL OF EPIDEMIOLOGY

occupational exposures, conditions in the residential area (local air pollution) and dietary habits. The section on food frequency contained 37 questions divided into four blocks and covered the intake of over 80 food items relevant for the Swedish population. All commonly consumed vegetables in Sweden were included in the questionnaire. The frequency questions were 'seldom or never', 'once or twice/month', 'once or twice/week', 'daily or almost daily' and, for commonly used food items, 'several times/day ... how many?' The respondent was encouraged to think about how he had eaten during the past 12-month period on average, and whether (and if so when) he changed eating habits drastically during the past 20 years (drastic change means that consumption changed over two frequency groups, or from 'daily' to 'several times daily)'. In the analysis, we have used the data for food intake 3 years prior to the interview. A vegetable index was formed by amalgamating the intake of carrot, tomato, cabbage, green pepper and lettuce. The consumption of each of these vegetables was weighted as 0 for 'seldom/never' or 'once/twice per month', 1 for 'once/twice per week' and 2 for 'daily/ almost daily consumption'. The sums were divided into three classes: 0-1 forming vegetable class 0, 2-4 class 1 and 5-10 class 2. A similar index was formed for fruits by adding the intake of citrus fruit, banana and 'other fruits and berries'. The frequencies of each of these fruits were weighted in the same way as the intake of vegetables. The sums were divided into three classes: 0-1 forming fruit class 0, 2-4 class 1 and 4-6 class 2. Interviews The interviews were performed by two nurses who had been employed and specially trained for the project. Patients who were reported to the department were as a rule contacted for interviews within a few days. Thus the interview could generally be conducted before the diagnosis was established or before the patient's condition had become so serious that an interview could not be carried out. The interview took about 45 minutes to complete and the two nurses alternated between interviewing controls and suspected cases in order to avoid subjective biases. Interviews with controls usually took place at the department or at their homes within 4—8 weeks of the patient interview. The recruitment of patients started in January 1989. There were breaks each summer between June and September as well as an interruption between May 1992 and February 1993. This paper described the analysis of all male cases of lung cancer and population controls

TABLE 1 Histological classes and basis of diagnosis among interviewed as well as not interviewed cases of lung cancer (ICD-7, 162.1) Interviewed

Not interviewed

No.

308(100)

36(100)

Histological class Small cell Squamous cell Adenocarcinoma Others

54(18) 133(43) 64(21) 57(19)

6(17) 18(50) 6(17) 6(17)

Basis of diagnosis X-ray diagnosis Histological examination of surgical or biopsy material Autopsy with histopathological examination of surgical or biopsy material Cytological diagnosis

1 (0.3)

0

198 (64)

16(44)

55(18) 54(18)

7(19) 13 (36)

interviewed between January 1989 and June 1993. The Boras General Hospital left the study in 1991 because of funding difficulties. Statistical Treatment of Data For estimation of odds ratios (OR), logistic regression models were fitted to the data with the EGRET software package for unconditional maximum likelihood estimation of the regression parameters. In all analyses, adjustments were made for age, number of cigarettes/day, number of years smoked (continuous variables), marital status (four classes) and socioeconomic job classification (seven classes). One gram of pipe tobacco was considered to be equal to one cigarette. The food frequency categories 'seldom or never' and 'once or twice/month' were combined into one reference category (low) in all analyses except for coffee, the reference for which was 'less than once or twice/week'. For vegetables and fruits, a consumption of 'once or twice/week' was classified as medium exposure, while 'daily/almost daily' was classified as high exposure. In addition, trend analyses with 2 x K tables were made after stratification for marital status, socioeconomic job classification, number of cigarettes/day (four classes) and number of years smoked (six classes).

RESULTS Table I shows the histological classes and basis of diagnosis among interviewed and not-interviewed lung cancer cases. Among the interviewed, 43% were squamous cell carcinomas, 21% adenocarcinomas, 18% small

35

DIETARY FACTORS AND LUNG CANCER

cell carcinomas and 19% others. Corresponding proportions among those not interviewed were 50%, 17%, 17% and 17%. The proportion of cases diagnosed by histological examination of surgical or biopsy material was higher among interviewed patients than among those not interviewed. The reverse relationship was seen for cases based upon cytological diagnosis. A breakdown of the characteristics of the patients and controls is given in Table 2. Most of the lung cancer cases were in the age group 60-69 years. Of the cases, 5% were younger than 50 years. As regards smoking history, 5% of the male cases were classified as never smokers. In contrast, 32% of the controls were never smokers. Significant differences between cases and controls were also seen for marital status, socioeconomic job classification, smoking duration and number of cigarettes smoked. Table 3 reports the proportion of high and low consumption of vegetables and fruits for cases and controls as well as OR for developing lung cancer with reference to low consumption (not more than once or twice/month). Differences between cases and controls were present at both ends of the frequency consumption scale for most kinds of vegetables and some fruits. Significantly lower OR were found for high consumption of carrots, cabbage, green pepper and 'other fruits and berries'. There were also significant trends for the OR for these vegetables as well as for tomatoes. With regard to the vegetable index, significant differences were seen between cases and controls; 34% of the cases and 17% of the controls were classified as low consumers, while 17% of the cases and 30% of the controls were high consumers. Adjusted OR for the vegetable index against class 0 (low consumption) demonstrated significantly lower risks; class I: OR = 0.69 (95% C I : 0.46-1.05), class 2: OR = 0.37 (95% CI : 0.23-0.61), P for trend < 0.001. When the analysis of vegetable index was restricted to current smokers only, the OR for class 1 was 0.56 (95% C I : 0.31-1.03) and 0.37 (95% C I : 0.18-0.76) for class 2. The protective effect of high vegetable consumption was most evident among those who had been smoking for at least 30 years: OR = 0.30 (95% C I : 0.17-0.53). For those with a smoking duration of less than 30 years and non-smokers, OR was 0.66 (95% CI : 0.24-1.82). Low OR for class 2 versus class 0 of the vegetable index were consistent for all histological types of lung cancer adenocarcinoma OR = 0.31 (95% CI: 0.14-0.72), small cell carcinoma OR = 0.47 (95% C I : 0.20-1.11), squamous cell carcinoma OR = 0.37 (95% CI: 0.19-0.71) and other types OR = 0.27 (95% CI : 0.10-0.69).

TABLE 2 Characteristics of lung cancer cases and population controls regarding age, marital status, socioeconomic job classification and smoking habits

No Age = 0.0071 Smoking history Never smoker Former smoker Smoker

Smoking duration (never smokers excluded) 1-19 years 20-29 30-39 50+

mean ± SD

16(5) 160(32) 98 (32) 214(42) 194 (63) 130(26) X2= 104.1, P-COOOOl

13(4) 84(24) 17(6) 64(19) 57(20) 71(21) 104(36) 85(25) 101(35) 40(12) X2» 105.53, P< 0.0001 43.6±U.O 31.9±15.l

Number of cigarettes smoked (never smokers excluded) 1-9 rig/day 10-19 20+ mean ± SD

78(15) 13(3) 44(9) 91 (18) 74(15) 117(23) 87 (17)

39(13) 96(28) 111(38) 122(35) 142(49) 126(37) = 21.43, P< 0.0001 18.5 ±10.1 15.9 ±11.7

36

INTERNATIONAL JOURNAL OF EPIDEMIOLOGY

TABLE 3 Number of lung cancer cases and controls who reported eating vegetables and fruit not more than once or twice/ month (low exposure), once or twice/week (medium exposure) and daily/almost daily (high exposure) respectively. Odds ratios are adjusted for age, number of cigarettes/day, number of years smoked, marital status and socioeconomic job classification Cases (n •» 308)

Controls (n = 504)

Adjusted odds ratio (95* confidence interval)

128 (42) 126(41) 54(18)

163(32) 198 (39) 143 (28)

1.0 0.93(0 63-1.36) 0 57 (0 36-0.90) P for trend = 0.009

263 (85) 9(3)

365 (72) 101 (20) 38(8)

1.0 0.64 (0.40-1 03) 0.37(0.17-0.83) P for trend = 0 010

Onion low medium high

117(38) 158(51) 33(11)

169(34) 277 (55) 58(12)

1.0 0 79 (0 55-1.12) 0.75 (0.43-1.32) P for trend = 0.071

Lettuce low medium high

232 (75) 56(18) 20(6)

316(63) 125 (25) 63(13)

1.0 0.88(0.58-1.35) 0.57 (0.31-1.04) /•for trend =0.126

Tomato low medium high

83 (27) 131 (43) 94 (31)

100 (20) 195 (39) 209(41)

1.0 1.10(0.71-1.71) 0.67(0.43-1.05) /•for trend = 0 027

Green pepper low medium high

229 (74) 60(20) 19(6)

320 (63) 126 (25) 58(12)

1.0 0.83 (0.55-1.25) 0.44(0.24-0.81) P for trend =0.007

Beans low medium high

271 (88) 34(11) 3(1)

433 (86) 65 (13) 6(1)

1.0 0.71 (0.43-1.16) 0.81 (0.16^.03) P for trend =0.058

Citrus fruit low medium high

92 (30) 95(31) 121 (39)

116(23) 174 (35) 214 (42)

1.0 0.81 (0.52-1.25) 0.90(0.59-1.36) Pfortrend =0.182

Rose hip soup low medium high

236 (77) 59(19) 13(4)

416 (83) 74 (15) 14(3)

1.0 1.44(0.91-2.26)

Carrol low medium high Cabbage low medium high

36(12)

P for trend =0.469 Banana low medium high

128(42) 114 (37) 66(21)

181 (36) 187 (37) 136 (27)

1.0 0.80(0.55-1.16) 0.79(0.51-1.21) P for trend =0.055 continued

TABLE 3 continued Cases (n = 308)

Controls (n = 504)

Adjusted odds ratio (95% confidence interval)

Other fruits or berries low medium hjgh

91 (30) 116(38) 101 (33)

77(15) 186(37) 241 (48)

1.0 0.72(0.47-1.12) 0.58 (0.37-0.89) P for trend =0.008

Vegetable index Class 0 Class 1 Class 2

105 (34) 150(49) 53(17)

85(17) 270 (54) 145 (30)

1.0 0.69(0.46-1.05) 0.37 (0.23-0.61) P for trend < 0.001

Fruit index Class 0 Class 1 Class 2

66(21) 180(58) 62 (20)

68(13) 300(60) 136 (27)

1.0 0.83(0.53-1.30) 0.73 (0.43-1 23) P for trend =0.014

The analysis of the fruit index, with adjustment for the same variables as in the analysis of vegetable index, did not show any strong protective effect of high consumption. The OR for fruit class 1 was 0.83 (95% CI : 0.53-1.30) and for class 2 OR = 0.73 (95% CI : 0.43-1.23). Those who had a high intake of fruits also had a high intake of vegetables. The OR for 'other fruits and berries' (Table 3) increased from 0.58 (95% CI : 0.37-0.89) to 0.74 (95% CI : 0.46-1.18)) when vegetable class was included in the regression model. This extension also changed the OR for fruit index: class 1 OR =1.00 (95% CI : 0.63-1.59), class 2 OR =1.02 (95% C I : 0.58-1.79). Table 4 demonstrates adjusted OR with relation to the consumption of different beverages. There was a strong tendency for a dose-related increase in lung cancer risk from drinking milk (P for trend = 0.07). The increased risk was of borderline significance for drinking milk several times/day; OR = 1.73 (95% CI : 1.003.01). Further analyses of milk drinking and lung cancer showed that the OR for the highest exposure class was 1.55 (95% CI : 0.60-3.99) for adenocarcinoma, while it was 1.31 (95% C I : 0.62-2.76) for squamous cell carcinoma, 1.67 (95% C I : 0.60-4.65) for small cell carcinoma and 2.91 (95% CI : 0.91-9.31) for others. There was no indication of an increased lung cancer risk from drinking tea. If the classes 'daily/almost daily' and 'several times/day' were combined the OR was 1.08 (95% C I : 0.73-1.61). Table 5 shows the OR for lung cancer in relation to consumption of meat, fish, eggs, cheese and cooking

37

DIETARY FACTORS AND LUNG CANCER

TABLE 4 Number of lung cancer cases and controls and odds ratios for lung cancer in relation to consumption of milk, sour milk, coffee, tea and herbal lea. Odds ratios are adjusted for number of cigarettes/day, number of years smoked, marital status, socioeconomic job classification, vegetable class and

Cases (n ™ 308)

Milk not more than once or twice/month once or twice/week daily/almost daily several times/day Sour milk not more than once or twice/month once or twice/week daily/almost daily Coffee not more than once or twice/week daily/almost daily 7-25 times/week >25 times/week

Herbal tea seldom or never once or twice/month at least once/week

31 (10)

78(15)

28(9) 105 (34) 141 (46)

70(14) 187 (37) 169(34)

0.91 (0.45-1.85) 1.37(0 79-2.39) 1.73(1.00-3.01)

174(56)

242 (48)

1.0

64(21) 70 (23)

109 (22) 153 (30)

1 09(0.71-1.68) 0.98(0.65-1.48)

13(4)

39(8)

26(8) 134(44) 135 (44)

86(17) 242 (48) 137(27)

1.0

1.0

0.94 (0.38-2.29) 1.16(0.53-2.52) 1.60(0.72-3.54)

191 (62)

270 (54)

1.0

43 (14) 62 (20) 12(4)

78(15) 119(23) 37(7)

0.92(0 57-1.50) 1.19(0.78-1.83) 0.74(0.33-1.64)

449 (89) 26(5) 29(6)

1.0

287 (93) 11 (4) 10(3)

(95% confidence interval)

Controls Adjusted odds ratio (n = 504) interval)

Tea

not more than once or twice/month once or twice/week daily/almost daily several times/day

TABLE 5 Odds ratios for lung cancer in relation to consumption of meat, fish, eggs, cheese, and cooking fat. The logistic regression model includes age, number of cigarettes/day, number of years smoked, marital status, socioeconomic job classification, vegetable class, 'other fruit or berries' and milk

Eggs daily/almost daily

1.25(0.72-2.15)

Margarine daily/almost daily several times/day

0.70(0.37-1.34) 1.11 (0.70-1.75)

Butter daily/almost daily several times/day

0.68 (0.22-2.07) 0.84(0.51-1.39)

Cheese daily/almost daily several times/day 1 Liver paste daily/several times/day

1.67(0.82-3.40) 1.89(0 94-3 80) 1.44(0.86-2.43)

Smoked or salted fish at least once/week

0.77(0.52-1.15)

Boiled or fried fish at least once/week

0.78(0.52-1.17)

Boiled or roasted meat at least once/week

0.98(0 70-1.39)

Fried meat, meatballs, etc. at least once/week

0 86(0.56-1.32)

Smoked or grilled meat, sausages at least once/week

0.82(0.57-1.17)

1 73 (0.75-3.97) 1.30(0.52-3.20)

fat. A slightly, although not significantly, elevated OR was found for high consumption of cheese. DISCUSSION Methods According to the cancer registry, 93 male lung cancer cases were diagnosed at the participating clinics in the study base during the first year of the study. Only six of these (6.5%) had visited one of the three participating clinics during the study period without having been identified as a suspected case. In addition, the rate of participation among the patients with suspected lung cancer was high (88%). The lung cancer cases in the study can thus be considered representative for the cases in the study base. The participation rate of population controls was also high and exceeded that of many previous studies.

people. The controls used in this study are thus likely to be representative for the study base. The food frequency questionnaire is a commonly used method of studying the dietary intake in a population. Studies on validity and reliability carried out by comparisons with diet records or serum analyses generally show that food frequency questionnaires can measure individual intake for a variety of nutrients.12 In studies on diet and cancer, past intake is more relevant than intake at the time of the study. Reports on the intake 20 years ago for each food item would probably be strongly influenced by the intake at the time of the study.13 We thus asked about intake during the most recent 12 months and drastic changes during the last 20 years for each food item. In spite of this, misclassification due to minor changes cannot be excluded.

38

INTERNATIONAL JOURNAL OF EPIDEMIOLOGY

All information on past dietary intake was collected by in-person interviews of cases and controls. We think that this information is more accurate than information from proxy respondents. It is likely that the proportion of men having very low intake of vegetables would have been smaller if information from wives had been used. Validation of information on past dietary intake is difficult, especially in case-control studies when the disease under study implies a short survival time. In this study, we accepted the reported frequencies without any attempt to validate the data from other sources. There is thus a potential risk of non-differential misclassification of exposure, but by using the extreme alternatives in the analyses we believe that this error has been minimized. Differential misclassification occurs if there is a difference between cases and controls in giving accurate answers to the questions on diet. We think that the relationship between dietary factors and lung cancer is not generally known by the population, in contrast to the situation for smoking. In addition, the patients were often interviewed before diagnosis was known, and recall bias thus seems unlikely. Results A major result from the study is that a high intake of vegetables was associated with a decreased risk of lung cancer among males. This protective effect of vegetables among males in in agreement with many previous studies. Block et al? reviewed the literature on diet and cancer and reported that all 17 case-control studies analysed demonstrated a protective role of fruit or vegetable consumption. The protective effect of vegetable intake was found for four different histological groups of lung cancer. This is in contrast to some other studies,14'13 where the protective effect was limited to squamous cell and small cell cancer, with no protective effect against adenocarcinoma. The study showed a protective effect of vegetables but not of fruits. The daily consumption of 'other fruit or berries', which in Sweden mainly comprises apples and pears, initially appeared to be protective against lung cancer. When the variable 'vegetable index' was included in the statistical model, however, the OR for 'other fruits and berries' increased to 0.76 and was no longer statistically significant. The highest class of the variable 'fruit index' showed a weak protective effect, but after inclusion of 'vegetable index' the point estimate of the OR for high fruit consumption was close to 1.0. The protective effect of vegetables but not fruits has previously been reported in studies from Hawaii15 and

Canada.17 In a study among non-smoking women in Missouri, USA, Alavanja et al.6 found a significantly protective effect of 'other fruit', which was no longer significant when other dietary factors were taken into consideration. Swanson et a/.18 reported, in a study from China, initially significantly decreased risks after high intake of bananas, apples, pears and some vegetables. After a factor analysis, the protective effect of vegetables remained, while the trend for fruit intake was inconsistent. Steinmetz et al}9 reported from the Iowa Women's Health Study a strong protective effect of vegetables but a weaker and non-significant effect of fruits. The questions about fruits were designed to evaluate the role of vitamin C. Rose hip soup was included because it is an item which is high in vitamin C and is commonly consumed in Sweden. No significant effects of food items containing vitamin C could be detected. The crude OR for citrus fruits of 0.71 increased to 0.90 after adjusting for smoking habits, marital status and socioeconomic job classification. Although intake of citrus fruits often has been suggested to be protective, a reverse relationship has recently been reported among females.6 A possible explanation for the lack of protective effect of high fruit intake may be that Swedish men in the studied age interval get their main vitamin C intake from sources other than fruits. In all, 85% of the men (similar for cases and controls) reported daily or almost daily intake of potatoes. This indicates that there are quite a few Swedish men with very low vitamin C intake. In this analysis, no attempt was made to transfer data on food items into nutrient values. Most retrospective and prospective studies suggest that vitamins, mainly carotenoids, are the specific protective agents, although no protective effect on lung cancer risk could be demonstrated in a recent intervention study on intake of beta-carotene and alpha-tocopherols.20 While these results may be explained by too short a duration of treatment, factors other than the pure vitamins might be of importance. From a public health point of view, the identification of relevant food items and advice on how often (or seldom) they should be consumed are quite adequate for prevention purposes. The only food item that was related to a significant increase in lung cancer risk was the consumption of milk. The risk increase seemed to be dose-related. This relationship was also found in previous studies in the USA21-22 and in India.23 The underlying cause could be the fat in the milk. Fat has been demonstrated to be a risk factor for lung cancer,4"6'24 and low fat milk has been related to a lower risk and even protection.21 In

DIETARY FACTORS AND LUNG CANCER

this study, the material was too small to allow for an evaluation of different kinds of milk. No association was seen between tea drinking and lung cancer in contrast to previous studies. One possible explanation is that only 4% of the cases and 7% of the controls drank tea more than once a day. In a British study,9 90% of the cases drank > 3 cups/day. In view of the results from milk consumption, another possible explanation might be that milk in the tea was a confounder in the British study. In conclusion, this study showed a strong protective effect of a high intake of vegetables in males and no pronounced effect of fruit intake. No increased risk could be demonstrated for tea drinking, while milk drinking was associated with a dose-related risk increase. The results demonstrate that diet is a potential confounder in studies on lung cancer and environmental factors. Diet should thus be taken into consideration in planning of such studies.

ACKNOWLEDGEMENTS This study was supported by the Swedish Cancer Foundation (grant 90-1137), the King Gustav V Jubilee Clinic Cancer Research Foundation, Gothenburg, Sweden, the Department of Community Medicine, Alvsborg County, and Forschungsgesellschaft Rauchen und Gesundheit MHB, Hamburg, Germany. The authors wish to thank Inga Molin, Karin Franzon and Anne-Marie Hilmersson for excellent work in data collection and Professor Hans Wedel for valuable discussions in planning of the study.

7

Goodman M T, Kolonel L N, Yoshizawa C N, Hankin J H. The effect of dietary cholesterol and fat on the risk of lung cancer in Hawaii. Am J Epidemiol 1988; 128: 1241-55. 8 Rylander R, Peterson Y, Lebowitz M. Assessing low-risk agents for lung cancer: methodological aspects. Int J Epidemiol 1990; 19 (Suppl. 1): S2-S86. 'Kinlen L J, Willows A N, Goldblatt P, Yudkin J. Tea consumption and cancer. Br J Cancer 1988; 58: 397-401. 10 Stocks P. Cancer mortality in relation to national consumption of cigarettes, solid fuel, tea and coffee. Br J Cancer 1970; 24: 215-25. "Tewes FJ, Koo L, Meisgen T J, Rylander R. Lung cancer risk and mutagenicity of tea. Env Res 1990; 52: 23-33. 12 Willett WC, Sampson L, Stampfer M J et al. Reproducibility and validity of a semiquantitative food frequency questionnaire. Am J Epidemiol 1985; 122: 51-65. 13 Mailer-Jensen O, Wahrendorf J, Rosenqvist A, Geser A. The reliability of questionnaire-derived historical dietary information and temporal stability of food habits in individuals. Am J Epidemiol 1984; 120: 281-90. 14 Ziegler P G, Mason T J, Stemhagen A et at. Carotenoid intake, vegetables and the risk of lung cancer among white men in New Jersey. Am J Epidemiol 1986; 123: 1080-93. 15 Byers T, Vena J, Mettlin C, Swanson M, Graham S. Dietary vitamin A, carotene, vitamin C and lung cancer risk: an analysis of histologic subtypes. Am J Epidemiol 1984; 120: 769-76. 16 he Marchand L, Yoshizawa C N, Kolonel L N, Hankin J H, Goodman M T. Vegetable consumption and lung cancer risk: A population-based case-control study in Hawaii. J Natl Cancer Insl 1989; 81: 1158-64. 17 Jain M, Burch J D, Howe G R, Risch H A, Miller A B. Dietary factors and risk of lung cancer: Results from a case-control study, Toronto, 1981-1985. Int J Cancer 1990; 45: 287-93. " Swanson C A, Mao B L, Li L Y et al. Dietary determinants of lung-cancer risk: Results from a case-control study in Yunnan province, China. Int J Cancer 1992; 50: 876-80. "Steinmetz K A, Potter J D, Folsom A R. Vegetables, fruit and lung cancer in the Iowa Women's Health Study. Cancer Res 1993; 53: 536-43. 20

REFERENCES 1 Bjelke E. Dietary vitamin A and human lung canter. Int J Cancer 1975; 15: 561-65 2 Fontham E. Protective dietary factors and lung cancer. Int J Epidemiol 1990; 19 (Suppl. 1): S32-S42. 3 Block G, Patterson B, Subar A. Fruit, vegetables, and cancer prevention: A review of the epidemiological evidence. Nutr Cancer 1992; 18: 1-29. 4 Wynder E L, Herbert J R, Kabat G C. Association of dietary fat and lung cancer. J Nail Cancer Insl 1987; 79: 631-37. 3 Wynder E L, Taioli E, Fujita Y. Ecologic study of lung cancer risk factors in the US and Japan, with special reference to smoking and diet. Japan J Cancer Res 1992; 8: 418-23. 6 Alavanja M C R, Brown C C, Swanson C, Brownson R C. Saturated fat intake and lung cancer risk among nonsmoking women in Missouri. J Natl Cancer Inst 1993; 85: 1906-16.

39

Heinonen O, Albanes D et al. (The Alpha-Tocopherol, Beta Carotene Cancer Prevention Study Group). T h e effect of vitamin E and beta carotene on the incidence of lung cancer and other cancers in male smokers. New Engl J Med 1994;

330: 1029-35. Mettlin C. Milk drinking, other beverage habits and lung cancer risks. Int J Cancer 1989; 43: 608-12. ^Maync ST. Janerich DT, Greenwald P et al. Dietary beta carotene and lung cancer risk in US nonsmokers. J Natl Cancer Inst 1994; 86: 33-38. 23 Sankaranarayanan R, Varghese C, Duffy S W, Padmakumary G, Day N E, Nair M K. A case-control study of diet and lung cancer in Kerala, South India. Int J Cancer 1994; 58: 644-49. 24 Shekelle R B, Rossof A H, Stamler J. Dietary cholesterol and incidence of lung cancer The Western Electric Study. Am J Epidemiol 1991; 134: 480-84. 21

(Revised version received June 1995)