Does insufficient adjustment for smoking explain the preventive effects of fruit and vegetables on lung cancer?

Does insufficient adjustment for smoking explain the preventive effects of fruit and vegetables on lung cancer?

Lung Cancer (2004) 45, 1—10 Does insufficient adjustment for smoking explain the preventive effects of fruit and vegetables on lung cancer? Halla Skul...

147KB Sizes 3 Downloads 111 Views

Lung Cancer (2004) 45, 1—10

Does insufficient adjustment for smoking explain the preventive effects of fruit and vegetables on lung cancer? Halla Skuladottir a,*, Anne Tjoenneland a , Kim Overvad b,c , Connie Stripp a , Jane Christensen a , Ole Raaschou-Nielsen a , Jørgen H. Olsen a a

Institute of Cancer Epidemiology, Danish Cancer Society, Strandboulevarden 49, 2100 Copenhagen, Denmark b Department of Clinical Epidemiology, Aalborg Hospital, Aarhus University Hospital, Aarhus, Denmark c Department of Epidemiology and Social Medicine, University of Aarhus, Aarhus, Denmark Received 18 July 2003 ; received in revised form 19 December 2003; accepted 29 December 2003

KEYWORDS Cohort study; Lung cancer; Diet; Smoking; Incidence

Summary Recent reports have raised the question, whether the previously observed protective effects of high intake of fruit and vegetables on the risk of lung cancer were due to insufficient adjustment for smoking leading to residual confounding. Association of intake of fruit and vegetables on lung cancer risk was examined, using the Danish prospective cohort study, ‘‘Diet, Cancer and Health’’. Participants completed a food-frequency and lifestyle questionnaire, and age-standardized incidence rates and rate ratios were estimated for quartiles of dietary exposure. In 1993—2001, 247 out of the 54 158 participants were diagnosed with lung cancer. The incidence rate of lung cancer was highest in the lowest quartile of intake of plant food (fruit, vegetables, legumes and potatoes) and the age-standardized rate ratio of lung cancer decreased significantly with increasing intake of plant food to 0.35 (95% CI, 0.27—0.45) but after control for smoking it was attenuated to 0.65 (95% CI, 0.45—0.93). The incidence rate differences of current smokers with high (≥400 g per day) and low (<400 g per day) daily intake of plant food were independent of smoking intensity; assuming a true biological protective effect, 80—90 lung cancer cases per 100 000 current smokers could be prevented in our cohort if all smokers had a high intake of plant food. The observed inverse association between high intakes of plant food seems chiefly to be a real protective effect, and not solely due to residual confounding. © 2004 Elsevier Ireland Ltd. All rights reserved.

1. Introduction Fruits and vegetables have been found to reduce the risk of many cancer types, among them lung * Corresponding author. Tel.: +45-35-25-76-55; fax: +45-35-25-77-34. E-mail address: [email protected] (H. Skuladottir).

cancer. Retrospective studies have reported relative risk estimates in the order of 0.5 for associations between high intake of specific vegetables, total fruit or vegetable intake and lung cancer risk, while prospective studies show mixed results with relative risk estimates between 0.7 and unity [1]. The protective effect has been reported among smokers, ex-smokers and never smokers

0169-5002/$ – see front matter © 2004 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.lungcan.2003.12.013

2 and for all histological subtypes of lung cancer [2]. In a recent report by Feskanich et al. however, the protective effect of high intake of fruit and vegetables on the risk of lung cancer was reduced substantially when control for self-reported tobacco exposure was improved step by step in the analyses of two large prospective cohorts [3]. That observation, together with the unexpected findings of three randomized trials of no reduction of beta-carotene supplementation on lung cancer risk [4—6]. raised the question whether the previously observed protective effects of high intake of fruits and vegetables on lung cancer observed in observational studies were due to insufficient adjustment for smoking, resulting in residual confounding. Further, a simulation study assuming a modest inverse relation between true tobacco exposure and serum beta-carotene suggested that biases in assessment of smoking exposure between smokers with low versus high beta-carotene intake, leading to residual confounding, may plausibly explain much or all of the observed protective effect of high beta-carotene levels seen in epidemiologic studies [7]. This study will examine the association between consumption of fruit and vegetables and lung cancer, using the Danish prospective cohort study, ‘‘Diet, Cancer and Health’’, exploring further the question whether the previous reported protective effects of high intake of fruit and vegetables, particularly seen in retrospective studies, could chiefly be explained by residual confounding by smoking.

2. Methods The prospective cohort study ‘‘Diet, Cancer and Health’’ was initiated by the Danish Cancer Society. Invited to participate were 80 996 men and 79 729 women aged 50—64 years. Invited participants were randomly selected from the Central Population Registry, which since 1 April 1968 has kept records on all permanent residents of Denmark with variables including full name and a unique 10 digit personal identification number. The personal identification number incorporates sex and date of birth and permits accurate linkage of information between registers. Eligible were persons born in Denmark, living in the greater Copenhagen or Aarhus areas, who according to the files of the Danish Cancer Registry at the time of invitation had no previous cancer diagnosis registered. In all, 27 178 men and 29 875 women were recruited during the time period December 1993—May 1997 (overall participation proportion 35%). All participants completed a detailed,

H. Skuladottir et al. validated 192-item food-frequency questionnaire, asking the participants to recall their dietary intakes over the past year, and a lifestyle questionnaire on known or suspected risk factors for cancer development such as current and past smoking habits, occupation, education, and previous health. The self-administered food frequency questionnaire was developed and validated before the study started [8,9]. The aim was to be able to rank the individuals with regard to intake of 19 different nutritients considered to be of prime importance in human carcinogenesis [8]. The validated food frequency questionnaire consisted of 175 questions with 92 questions on different foods and recipes. There were separate questions on portion sizes. For food items used on bread like cold meat and cheese, separate questions were asked for their use on rye bread and white bread, respectively. Since food frequency questionnaires tend to overestimate the consumption of single food items, the questionnaire was designed with a number of global questions to check the frequency of consumption of fruit and vegetables, amongst other major food items, allowing for an adjustment for overestimation of consumption. The final food frequency questionnaire used in the ‘‘Diet, Cancer and Health’’ study was altered according to the experience from the validation study and included 192 items and is therefore not identical with the food frequency questionnaire used in the validation study. In all 31 varieties of vegetables (excluding potatoes and vegetable juices) and 16 varieties of fruit (excluding fruit juices) were assessed. The questionnaire was optically scanned after completion and checked for any missing or inconsistent data by the study personnel, and if found, was corrected in a subsequent personal interview. In this study we constructed the following groups of foods: (1) vegetables, containing all vegetables excluding potatoes and vegetable juices; (2) fruits, containing all fruits, excluding fruit juices; and (3) plant foods, comprising all vegetables and all fruit, additionally including vegetable and fruit juices, legumes and potatoes. In January 2002, all 57 053 members of the study cohort were re-linked to the Central Population Register for information on vital status and migration. Information on cancer occurrence till the end of 2001 was obtained through record linkage to the Danish Cancer Registry that collects information on all patients with cancer in Denmark. The Cancer Registry was initiated in 1942, and physicians provided data voluntarily to the Registry until 1987, when notification of malignant and related diseases was legislated. A high degree of completeness in cancer registration has been achieved by a system

Effects of fruit and vegetables on lung cancer of multiple sources for certification: (1) reports from clinical departments on newly diagnosed cases; (2) reports from pathology departments on all malignancies discovered, including cancers first recognized at autopsy; and (3) the investigation of cases based on death certificates only. Since 1987 the cancer registration has been supplemented with patients identified by linkage to the National Hospital Discharge Registry [10]. From the study cohort, we excluded 537 persons who were later reported to the Danish Cancer Registry with a cancer diagnosed prior to entry into the study, and 51 and 2307 persons with incomplete or inconsistent data on dietary intake or smoking habits, respectively. The remaining 54 158 cohort members (94.9%) were followed for lung cancer occurrence from the date of entry into the study until the date of death (n, 1045), date of emigration (n, 181), or 31 December 2001.

2.1. Statistical analysis Age-standardized lung cancer incidence rates were calculated for men and women, separately, for employment in risk occupations and for various levels of smoking and fruit, vegetable and plant food consumption using direct standardization to the age distribution of the entire study cohort in 5 years categories. Associated standardized incidence rate ratios, SIRs–—a measure of the relative risk–—were derived, and two sided 95% confidence intervals were calculated. Smoking status was defined at the entry by the participant’s responses to two questions, whether they ever had smoked daily and whether they currently smoked daily. Indicators for smoking habits (current, former or never smoker), current number of cigarettes smoked per day, duration of smoking in years and an indicator for ever being occupied in an industry or job associated with higher risk of developing lung cancer for at least 1 year, were considered as potential confounders for lung cancer. The following industries or jobs were considered to be associated with higher risk of lung cancer: asbestos, building, cement, chemical, glass, and graphic industry, leather tannery, metal processing, electroplating, foundry, shipbuilding, rubber, and textile industry, transport, auto mechanic, cook, petrol pump attendant and waiter. Cox proportional hazards model including time dependent variables was used to control for potential confounders of the association between the amount of intake of fruit, vegetables, plant food and lung cancer, using the PHREG procedure in SAS 6.12 (SAS Institute, Inc., Cary, NC, USA). Age was used as time scale to ensure that the estimation

3 procedure was based on comparisons of individuals of the same age. Time under study was always included as a time dependent variable and was modeled by a linear spline with a boundary at 1 year after entry into the study cohort. Instead of excluding the cases occurring the first year after entry into the cohort, we allowed the incidence rate of lung cancer to change with time since baseline using time-dependent variables to model a linear spline, with a boundary at 1 year after entry into the study cohort [11]. Analyses were conducted for men and women combined, assuming the same effect of known risk factors of lung cancer in both genders, and adjustment for smoking and occupational exposure to lung carcinogens was performed stepwise in order to examine the resulting effects on the strength of the association. In the last model though, we performed separate analyses for men and women to test our assumption of same effect in both genders and found that gender did not influence the estimates. Quartiles of intake of fruit, vegetables and plant food were formed on the basis of the frequency distribution among lung cancer cases to fully utilize the statistical power of the study. In the fully adjusted model we also estimated the rate ratio for lung cancer per 100 g increment of the daily intake of fruit, vegetables or plant food; the quartile categories used in previous analyses was disregarded, and intake of fruit, vegetables and plant food was entered as a continuous variable, estimating the trend and thus providing analyses of linear trend. In the initial analyses, each of the continuous variables, including the previously mentioned confounders, was modeled in the Cox regression model as a linear spline, and the hypothesis of a linear association was tested. We found no significant systematic departure from linearity. Similar analyses were performed for following histological subtypes of lung cancer: adenocarcinomas (n = 80), squamous cell carcinomas (n = 50), small cell carcinomas (n = 44) and the rest (n = 73), the latter of which also included 10 cases of lung cancer with no histological data. Two-sided 95% confidence intervals for the rate ratios were calculated on the basis of Wald’s test of the Cox regression parameter, that is, on the log rate ratio scale. In order to evaluate whether the level of smoking modifies the protective effect of high intake of plant food (≥400 g per day, roughly corresponding to the two highest quartiles of intake) on the incidence rate of lung cancer we analyzed the protective effect (measured as preventable lung cancers per 100 000 subjects per year) in current moderate smokers and current heavy smokers, respectively. Moderate smokers comprised of cohort members

4

H. Skuladottir et al.

smoking at baseline, with self-reported cigarette consumption since age 20 corresponding to less than 20 pack-years of cigarette smoking, and heavy smokers by members with self-reported cigarette consumption since age 20 corresponding to ≥20 pack-years of cigarette smoking. Pack-years were defined as mean number of cigarettes smoked per day reported in 10 years intervals from 20 years of age to age at stop or age at entry, multiplied with years of duration of smoking (subtracting the numbers of years with smoking pauses, if any) and divided by 20.

3. Results A total of 247 study participants (142 men and 105 women) were diagnosed with lung cancer between the date of entry and the end of 2001 (Table 1). The crude incidence rate of lung cancer was higher among men (143.3 per 100 000 per year)

than among women (94.6 per 100 000 per year) and was higher in the older age groups − 68.4 per 100 000 per year in the 50—54-year-old and 206.8 in the 60—65-year-old. The age-standardized incidence rate was also markedly different according to smoking status, from 9.3 per 100 000 per year in never smokers to 64.1 in ex-smokers and 272.7 per 100 000 per year in current smokers. Duration of smoking and number of cigarettes smoked per day was also strongly associated with the incidence rates of lung cancer; higher with longer duration of smoking and with greater amount of cigarettes smoked per day (Table 1). The age-standardized rate ratio (SIR) for lung cancer of persons having smoked for 1—19 years as compared to persons having smoked for 40—59 years was 0.05, and 0.26 for persons having smoked for 20—39 years. Similarly, people who at an average smoked less than 15 cigarettes per day had an SIR of 0.28 when compared to those who smoked over 25 cigarettes per day. Having ever been employed in a occupation

Table 1 Number of cases with lung cancer, person-years, incidence and age-standardized rate ratios according to different values of confounding variables for the Danish ‘‘Diet, Cancer and Health’’ study, 1993—2001 Cases (n = 247)

Personyears

Crude incidence rate per 100 000

Age-standardized incidence rate per 100 000

Age-adjusted rate ratio with 95% confidence intervala

Gender Men Women

142 105

99085 110970

143.3 94.6

144.9 93.8

1.53 (1.29—1.81) 1.00

Age (years) 50—54 55—59 60—65

61 71 115

89213 65225 55616

68.4 108.9 206.8

— — —

0.33 (0.25—0.43)b 0.53 (0.41—0.66)b 1.00

Smoking status (%) Never smoker Ex-smoker Current smoker

7 39 201

76084 59357 74614

9.2 65.7 269.4

9.3 64.1 272.7

0.03 (0.01—0.07) 0.24 (0.17—0.33) 1.00

7 87 146

30160 71907 31903

23.2 121.0 457.6

23.8 122.6 465.1

0.05 (0.02—0.10) 0.26 (0.21—0.32) 1.00

45 83 41

26695 29289 7794

168.6 283.4 526.0

159.6 297.8 601.1

0.28 (0.21—0.38) 0.47 (0.38—0.59) 1.00

Ever occupied in risk occupation Yes 134 No 113

72202 137853

185.6 82.0

187.7 81.5

2.27 (1.90—2.69) 1.00

Confounder variable

Duration of smoking (%) 1—19 years 20—39 years 40—59 years Cigarettes smoked per dayc ≤14 cigarettes per day 15—24 cigarettes per day ≥25 cigarettes per day

a

The highest category is used as the reference category. Crude rate ratios. c Seventy-one patients preferred other tobacco products (cigarillos, cigar or pipe) and seven were non-smokers. b

Effects of fruit and vegetables on lung cancer

Table 2 Median intake of plant food (g per day)

by smoking status in the Danish ‘‘Diet, Cancer and Health’’ study, 1993—2001

All fruits All vegetables Total plant food

Never smoker

Ex-smoker

Current smoker

164.9 169.4 522.2

148.7 169.2 516.4

109.0 138.0 449.6

associated with higher risk for lung cancer had an effect on the incidence rate of lung cancer, and the SIR was 2.27 when compared with cohort members without such employments, but after adjusting for smoking status, duration of smoking and number of cigarettes smoked per day the rate ratio decreased to 1.62 (95% CI, 1.26—2.09). Smoking is a confounder because it is associated with both lung cancer and the intake of plant food. Increasing smoking status category, ordered from never smoker over ex-smoker to current smoker was associated with less intake of fruit, vegetables and all plant food in general (Table 2). The most frequently consumed vegetables amongst cohort members were carrots, tomatoes and onion, and the most frequently consumed fruits were apples, pears and oranges (not shown). Table 3 shows the number of lung cancer cases, person-years at risk, age-standardized incidence rates and associated SIRs of lung cancer by quartiles of intake of fruit, vegetables and all types of plant food combined. The age-standardized incidence rate of lung cancer was clearly highest in the lowest quartile of intake of fruit and vegetables and lower in every succeeding quartile of intake. For fruit, the age-standardized incidence rate was 206.9 per 100 000 per year in the lowest quartile of daily intake, compared to 70.6 per 100 000 in the highest quartile. Similarly, for vegetables the age-standardized incidence rate was 228.2 per 100 000 in the lowest quartile of intake and 68.8 in the highest quartile. Thus, monotonic dose— response associations were observed in all three food groups in every succeeding quartile of intake. The SIRs and further adjusted rate ratios of lung cancer among cohort members with daily intake of fruit, vegetables and plant food above the lowest intake quartile (reference category) was estimated in models with gradually improved control for smoking habits and occupation (Table 3). In the first model with adjustment for age only, the SIRs of lung cancer among cohort members with intakes in the highest quartile was for fruit 0.35 (95% CI, 0.27—0.45), for vegetables 0.29 (95% CI, 0.22—0.37), and for all plant food 0.35 (95% CI,

5 0.27—0.45). In the final model, adjusting for age, smoking status, duration of smoking, number of cigarettes smoked per day and occupation, the seemingly protective effect of fruit and vegetables, respectively, was not anymore so clear. The protective effect of the overall intake of plant food was statistically significant, but weaker than before with a rate ratio in the highest quartile of 0.65 (95% CI, 0.46—0.94), corresponding to a 35% lower risk of lung cancer in subjects in the highest quartile of intake as compared to subjects in the lowest quartile of intake. The incidence rate ratios for increments of 100 g per day in intake of fruits, vegetables and plant food are given in Table 4. In the fully adjusted model, our findings indicate a decline in the risk of lung cancer in the order of 7% for each 100 g increment in the daily intake of all types of plant food combined. The associations observed for fruit or vegetables were in the expected direction but not statistically significant. Analyses on the three major histological subtypes of lung cancer showed a significantly protective effect of increments of 100 g per day of vegetables on the risk of small cell lung cancer, however, with a tendency for all incidence rate ratios to be 5—20% lower per 100 g per day at higher intake. To see whether the protective effects of high intake of fruit and vegetables were the same regardless of the level of smoking, we stratified the current smokers in the cohort approximating their cumulative amount of smoking; moderate smokers having smoked less than 20 pack-years and heavy smokers having smoked 20 pack-years or more, and in each subgroup we analyzed the incidence rate of lung cancer among individuals with high and low intake of all plant food (more than or equal to 400 g per day; less than 400 g per day). The median age in both groups of moderate smokers was 55 years, and the median pack-years of moderate smokers was 11.3 in participants with low intake and 10.5 in participants with high intake of plant food. Among heavy smokers the median age was 56 years in both groups, and the median pack-years of heavy smokers were 32.0 and 30.0 in participants with low or high intake of plant food, respectively. The incidence rate of lung cancer was higher in smokers with low intake of plant food in both smoking strata (Fig. 1). The incidence rate difference, often used as an estimate of the absolute impact of an association, was similar in both smoking strata; 90.2 per 100 000 per year in moderate smokers and 79.1 per 100 000 subjects in heavy smokers. Assuming a true biological protective effect, these findings indicate that in this cohort of middle-aged to old smokers, 80—90 lung cancer cases per 100 000 current

6

Table 3 Frequency, age-standardized incidence (SIR) and rate ratio of lung cancer by quartile of intake of total fruit and vegetable consumption with increasing detail in adjustment for smoking variables and 95% confidence intervals (CI) in the Danish ‘‘Diet, Cancer and Health’’ study, 1993—2001 Food group by quartiles

Number of cases

Personyears

Age-standardized incidence ratea

Age-standardized rate ratio (SIR)

Rate ratio model 2b

Rate ratio model 3c

Rate ratio model 4d

Fully adjusted rate ratioe

Fruit (g) 5—40 41—88 89—164 165—643

61 62 62 62

30653 36254 57148 86000

206.9 174.2 108.5 70.6

1.00 0.83 (0.63—1.06) 0.52 (0.40—0.67) 0.35 (0.27—0.45)

1.00 1.11 0.85 0.69

1.00 1.12 0.88 0.73

1.00 1.21 0.98 0.83

1.00 1.23 (0.86—1.76) 1.00 (0.69—1.44) 0.86 (0.59—1.26)

Vegetables (g) 17—68 69—120 121—170 171—479

62 62 62 61

25418 45730 46837 92070

228.2 132.0 132.8 68.8

1.00 0.60 (0.46—0.76) 0.58 (0.44—0.74) 0.29 (0.22—0.37)

1.00 0.73 0.89 0.57

1.00 0.74 0.92 0.60

1.00 0.77 0.96 0.64

1.00 0.78 (0.54—1.11) 0.99 (0.68—1.42) 0.67 (0.46—0.97)

1.00 0.84 0.66 0.65

1.00 0.84 (0.59—1.20) 0.67 (0.46—0.95) 0.65 (0.46—0.94)

H. Skuladottir et al.

All plant food (g) 78—291 61 27256 227.1 1.00 1.00 1.00 292—407 63 43236 148.0 0.64 (0.49—0.82) 0.78 0.79 408—566 61 62572 97.7 0.43 (0.32—0.55) 0.59 0.60 567—1394 62 76990 79.2 0.35 (0.27—0.45) 0.53 0.57 a Age-standardized incidence rate per 100 000. b Adjusted for age and smoking status. c Adjusted for age, smoking status and duration of smoking. d Adjusted for age, smoking status, duration of smoking and number of cigarettes smoked per day. e Adjusted for age, smoking status, duration of smoking, number of cigarettes smoked per day and occupation.

Effects of fruit and vegetables on lung cancer

7

Table 4 Adjusted rate ratios of lung cancer by histology per increase of 100 g per day of average daily quantity

of intake of fruit and vegetable with 95% confidence intervals (CI) in the Danish ‘‘Diet, Cancer and Health’’ study, 1993—2001

Histology

Food group

Age-adjusted rate ratio

Fully adjusted rate ratioa

All lung tumors combined (n = 247)

Fruit Vegetables All plant food

0.79 (0.70—0.89) 0.72 (0.61—0.85) 0.84 (0.79—0.89)

0.95 (0.86—1.06) 0.88 (0.75—1.03) 0.93 (0.88—0.99)

Small cell carcinoma (n = 43)

Fruit Vegetables All plant food

0.78 (0.58—1.06) 0.46 (0.29—0.74) 0.77 (0.65—0.90)

1.00 (0.77—1.30) 0.61 (0.39—0.97) 0.89 (0.77—1.04)

Adenocarcinoma (n = 79)

Fruit Vegetables All plant food

0.81 (0.66—1.00) 0.79 (0.60—1.05) 0.84 (0.74—0.93)

0.96 (0.80—1.15) 0.94 (0.72—1.23) 0.92 (0.82—1.02)

Squamous cell carcinoma (n = 49)

Fruit Vegetables All plant food

0.66 (0.49—0.91) 0.77 (0.53—1.12) 0.83 (0.72—0.95)

0.82 (0.62—1.09) 0.91 (0.63—1.30) 0.91 (0.79—1.04)

a

Adjusted for age, smoking status, duration of smoking, number of cigarettes smoked per day and occupation.

smokers per year could be prevented if all current smokers had a daily intake of 400 g or more of plant food per day.

4. Discussion In this prospective cohort study ‘‘Diet, Cancer and Health’’, includes 54 158 cohort members,

contributing 210 055 person years of observation, there were diagnosed 247 cases of lung cancer. We found that high intake of fruit, vegetables or total plant food had a significant inverse association with lung cancer but after stepwise adjustment for smoking status, duration of smoking and amount of cigarettes smoked per day the associations were attenuated and remained only significant for people in the highest quartile of intake of vegetables

Fig. 1 Incidence of lung cancer by pack-years and daily intake of plant food in the Danish ‘‘Diet, Cancer and Health’’ study, 1993—2001.

8 and all plant food; they had 35% lower risk of lung cancer as compared to those who were in the lowest quartile of intake. The incidence rate of lung cancer increased with age in the cohort as expected and reflected accurately the incidence rates in the corresponding age groups in the general population in Denmark in the period 1993—1996, indicating a good representation of the middle-aged and older Danish general population in the cohort [12]. The incidence rate of lung cancer varied markedly according to smoking status and the risk ratios for never smokers and ex-smokers using current smokers as a reference, correspond closely to the risk ratios given for never-, ex- and current smokers in a recent publication by Peto et al. [13]. Duration of smoking and number of cigarettes smoked per day also has profound effect on the incidence rate of lung cancer, corresponding closely to published reports. The association between intake of fruit, vegetables and lung cancer risk has been studied in several prospective cohort studies. Most risk estimates, although not significant, point to the same direction, towards a protective effect. Studies on the association of intake of fruit and risk of lung cancer have been partly inconsistent; some have found an inverse association [14—16] while other have not been able to reproduce these findings [17—23]. Similarly, studies on intake of vegetables and lung cancer risk have found insignificant or no association [14,15,18—20,22,23] and some have found a significant inverse association [3,16,17,24]. Imperfect assessment of the underlying true preventive factors, that still remain unidentified, may be the cause of fluctuations in the strength of association in different studies. Smoking is a critical exposure that may confound any analysis of the association between diet and lung cancer and may have been measured with variable accuracy in different studies, possibly contributing to the inconsistencies in their results. In our analysis we have adjusted for the potential confounding effect of smoking status, duration of smoking and amount of cigarettes smoked per day. We added smoking variables stepwise to our multivariate model to observe the effects of smoking on our estimates and observed an attenuation of our estimates with every added smoking variable. It is not possible to measure smoking accurately with current questionnaire methods; some error will always be present in any questionnaire survey data set, therefore some residual confounding will probably always be present, especially when the study exposure is weak compared with the confounder, as is the case with dietary exposure and cigarette smoking. Analyzing the effect of diet in non-smoking lung cancer patients

H. Skuladottir et al. minimizes the effect of confounding by smoking, although there still will remain a possible source of confounding in measurement of exposure to environmental tobacco smoke, although the confounding would be expected to be of small degree. Recently a large multicenter case-control study of diet and lung cancer among non-smokers was published, including 506 non-smoking lung cancer cases and 1045 non-smoking controls [25]. The authors found a significant protective effect of high intake of fresh vegetables (odds ratio = 0.7) and a non-significant protective effect of high intake of fruits. Previous observational studies (retrospective and prospective) either conducted extensively among non-smokers or having sufficient samples to analyze data for non-smokers separately, have found similar associations (reviewed in [26]) indicating a protective effect of fruit and vegetables in non-smokers. Due to these findings we believe that the observed inverse association between high intake of fruits and vegetables in the present study is chiefly a real protective effect, but not solely due to residual confounding. It still remains an unresolved question by what mechanism fruit and vegetables exert their protective effects and which specific factors in fruit and vegetables may be involved. If the protective effect of high intake of fruit and vegetables interfered with some processes involved in the harmful effect of tobacco smoke in vivo, one would assume that the protection would be quantitatively larger in heavy smokers as compared to light smokers. However, the extent of protection seemed to be equally large in moderate smokers and heavy smokers, suggesting a protective effect operating independently of smoking intensity. We categorized the lung tumors into histological subgroups and the distribution of the subgroups correspond well with previously published report of lung cancer in Denmark [12]. Few prospective cohort studies have analyzed the effect of fruit and vegetables on different histological subgroups of lung tumors [3,14,17,23,24]. Most studies have divided the lung tumors into two groups; Kreyberg I (comprising small cell carcinomas, squamous cell carcinomas and large cell carcinomas) and Kreyberg II (adenocarcinoma). Three studies found a stronger protective effect of high intake of fruit and vegetables in Kreyberg I tumors [3,23,24], one found a similar effect on both groups [14] and the last study found a large protective effect on large cell carcinomas [17]. In this study the protective effect was similar in all three histological groups, but higher intake of vegetables reduced the risk of small cell lung carcinoma significantly and to a greater extent than in the other histological groups. The

Effects of fruit and vegetables on lung cancer latter finding is possibly a chance finding but might be viewed in part consistent with the findings of previous studies, because small cell carcinoma is included in the Kreyberg I group. The questionnaire used in this study was validated against two times 7 days of weighted diet records and it was designed with a number of global questions to check the frequency of consumption, allowing for an adjustment for overestimation of consumption. When we look at the variation in diet, we find approximately 50% increase in intake between quartiles of intake. The National Health Institute in Denmark recommends an intake of 600 g per day of fruit and vegetables combined. Participants in the highest quartile of intake reach that recommendation. The intake of fruit and vegetables in the Danish population in general is very similar to the distribution of intake of fruits and vegetables in the cohort [27]. The observed effect of intake of fruit and vegetables on the risk of lung cancer might therefore to some extent be generalized to the middle-aged or older general population in Denmark. In conclusion, we found an inverse association between lung cancer risk and high intake of fruit, vegetables and total plant food. Stepwise adjustment for smoking status, duration of smoking and amount of cigarettes smoked per day attenuated the associations. Subjects in the highest quartile of intake of vegetables and all plant food had 35% lower risk of lung cancer as compared to those who were in the lowest quartile of intake. The beneficial effect of high intake of fruit and vegetables on lung cancer protection seems to be independent of smoking intensity. In our cohort of middle age or older smokers, 80—90 cases of lung cancer per 100 000 current smokers per year could be prevented if they all had a high intake of plant food, independently of their smoking intensity. The beneficial effect is small in comparison with the beneficial effects of stopping the habit of smoking, but important. We should continue to encourage smokers to give up the habit of smoking and to increase their intake of fruit and vegetables in order to reduce their risk of developing lung cancer.

9

[4]

[5]

[6]

[7]

[8]

[9] [10] [11] [12] [13]

[14] [15] [16] [17] [18] [19] [20]

References [1] World Cancer Research Fund. Food, nutrition and the prevention of cancer: a global perspective. Washington: World Cancer Research Fund in Association with American Institute for Cancer Research, 1997. [2] Ziegler RG, Mayne ST, Swanson CA. Nutrition and lung cancer. Cancer Causes Contr 1996;7:157—77. [3] Feskanich D, Ziegler RG, Michaud DS, Giovannucci EL, Speizer FE, Willett WC, et al. Prospective study of fruit

[21] [22] [23]

and vegetable consumption and risk of lung cancer among men and women. J Nat Cancer Inst 2000;92:1812—23. The Alpha-Tochopherol, Beta Carotene Cancer Prevention Study Group. The effect of vitamin E and beta carotene on the incidence of lung cancer and other cancers in male smokers. N Engl J Med 1994;330:1029—35. Omenn GS, Goodman GE, Thornquist MD, Balmes J, Cullen MR, Glass A, et al. Effects of combination of beta carotene and vitamin A on lung cancer and cardiovascular disease. N Engl J Med 1996;334:1150—5. Hennekens CH, Buring JH, Manson JE, Stampfer M, Rosner B, Cook NR, et al. Lack of effect of long-term supplementation with beta carotene on the incidence of malignant neoplasms and cardiovascular disease. N Engl J Med 1996;334:1145—59. Stram DO, Huberman M, Wu AH. Is residual confounding a reasonable explanation for the apparent protective effects of beta carotene found in epidemiologic studies of lung cancer in smokers. Am J Epidemiol 2002;155:622—8. Overvad K, Tjønneland A, Haraldsdottir J. Development of a semiquantative food frequency questionnaire to assess food, energy and nutrient intake in Denmark. Int J Epidemiol 1991;20:900—5. Tjønneland A, Overvad K, Haraldsdottir J. Validation of a semiquantitive food frequency questionnaire developed in Denmark. Int J Epidemiol 1991;20:906—12. Cancer Incidence in Denmark 1998. Sundhedsstatistikken 2002:1, Sundhedsstyrelsen. http://www.sst.dk/publ/Publ 2002/cancer%20incidence 98.pdf. Greenland S. Dose—response and trend analysis in epidemiology: alternatives to categorical analysis. Epidemiology 1995;6:356—565. Skuladottir H, Olsen JH, Hirsch FR. Incidence of lung cancer in Denmark: historical and actual status. Lung Cancer 2000;27:107—18. Peto R, Darby S, Deo H. Smoking, smoking cessation, and lung cancer in the UK since 1950: combination of national statistics with two case-control studies. Br Med J 2000;321:323—9. Fraser GE, Beeson WL, Philips RL. Diet and lung cancer in Californian Seventh-day Adventists. Am J Epidemiol 1991;133:683—93. Knekt P, Järvinen R, Seppänen R, Rissanen A, Aromaa A, Heinonen OP, et al. Dietary antioxidants and the risk of lung cancer. Am J Epidemiol 1991;134:417—79. Speizer FE, Coldiz GA, Hunter DJ. Prospective study of smoking, antioxidant intake, and lung cancer in middleaged women (USA). Cancer Causes Contr 1999;10:475—82. Kvåle G, Bjelke E, Gart JJ. Dietary habits and lung cancer risk. Int J Cancer 1983;31:397—405. Long-de Hammond EC. Lung cancer fruit, green salad and vitamin pills. Chin Med J 1985;98:206—10. Chow W-H, Schuman LM, McLaughling JK. A cohort study of tobacco use, diet, occupation and lung cancer mortality. Cancer Causes Contr 1992;3:247—54. Shibata A, Paganini-Hill A, Ross RK, Yu MC, Henderson BE. Dietary beta carotene cigarette smoking and lung cancer in men. Cancer Causes Contr 1992;3:207—13. Steinmetz KA, Potter JD, Folsom AR. Vegetables fruit, and lung cancer in the Iowa Women’s Health Study. Cancer Res 1993;53:536—43. Breslow RA, Graubard BI, Sinha R, Subar AF. Diet and lung cancer mortality: a national health interview survey cohort study. Cancer Causes Contr 2000;11:419—31. Voorrips LE, Goldbohm RA, Verhoeven DTH, van Poppel GA, Sturmans F, Hermus RJ, et al. Vegetable and fruit consumption and lung cancer risk in the Netherlands Cohort Study

10 on Diet and Cancer. Cancer Causes Control 2000;11:101— 15. [24] Hirayama T. Contribution of a long-term prospective cohort study to the issue of nutrition and cancer with special reference to the role of alcohol drinking. Prog Clin Biol Res 1990;346:179—87. [25] Brennan P, Fortes C, Butler J, Agudo A, Benhamou S, Darby S, et al. A multicenter case-control study of diet and

H. Skuladottir et al. lung cancer among non-smokers. Cancer Causes Control 2000;11:49—58. [26] Brownson RC, Alavanja MCR, Caporaso N, Simoes EJ, Chang JC. Epidemiology and prevention of lung cancers in nonsmokers. Epidemiol Rev 1998;20:218— 36. [27] Danskernes Kostvaner. Hovedresultater, Levnedsmiddelstyrelsen, 1995.