Excess body weight, as measured by body mass index (BMI; weight [kg] divided by height squared [m2]) is an important predictor of total mortality, cardiovascular disease mortality, and mortality from certain cancers.1 A recent analysis of a large cohort of retirees indicated that even moderate excess body weight in middle age is associated with increased total mortality.2 In apparent contrast to these deleterious effects of overweight and obesity, a number of epidemiologic studies, (both case-control and cohort designs) have reported an inverse association of body mass index with lung cancer risk, suggesting that leanness may be a risk factor for the development of lung cancer.3–13 However, other studies have either failed to find an association,14–16 or the association disappeared when smoking habits and health status were taken into account.17,18 One study in never-smokers and long-term ex-smokers reported a positive association between body mass index and lung cancer risk.19
The observed association between leanness and lung cancer risk may be explained by the fact that cigarette smokers tend to be leaner than nonsmokers; by weight loss due to pre-existing disease; by competing causes of death; or by some combination of these.18,20 However, several studies have at least partially addressed these concerns.3,4,7–9,11 Furthermore, although current smokers tend to weigh less than former smokers and never-smokers, BMI among current smokers tends to be higher in heavy smokers than in those smoking less.21–28 Thus, it is unclear that the inverse association between BMI and lung cancer risk can be explained by smoking. In view of the conflicting evidence, studies capable of addressing sources of bias or confounding can help determine whether there is a reproducible association between BMI and lung cancer. If BMI is truly associated with lung cancer risk, a second step would be to explain the biologic basis of the association.
We report here on the association between BMI and risk of lung cancer in a large prospective cohort of women followed for an average of 16 years.
The study, which has been described in detail elsewhere,29,30 was conducted among participants in the Canadian National Breast Screening Study, a randomized controlled trial of screening for breast cancer. A total of 89,835 women ages 40–59 years with no history of breast cancer were recruited into the trial between 1980 and 1985. The study was approved by the appropriate Institutional Review Boards, and informed consent was obtained from all study participants.
At recruitment into the study, participants completed a self-administered questionnaire that sought information on demographic characteristics, smoking history, menstrual and reproductive history, and use of oral contraceptives and replacement estrogens. Information about smoking habits included smoking status, number of cigarettes smoked per day, number of years of smoking, and, for former smokers, the year they had stopped smoking. Height and weight were measured at the initial physical examination. This information was available on the full cohort of 89,835 women. Starting in 1982 (that is, after some participants had completed their scheduled visits to the screening centers), a self-administered food-frequency questionnaire containing questions on 86 food items was distributed to all new attendees at all screening centers and to women returning to the screening centers for rescreening. This questionnaire also included questions on current weight, weight 1 year prior to enrollment, weight at age 20, height, alcohol consumption, and physical activity. A total of 49,654 women completed the food frequency questionnaire.
BMI was defined as weight (kg)/height (m2), (either as measured or self-reported). Pack-years of smoking were computed by multiplying the number of cigarettes smoked per day by the total years of smoking, and dividing by 20. Change in weight was computed by subtracting weight at age 20 (ie, self-reported weight in the subgroup with dietary information) from weight measured at enrollment. Data from the food frequency questionnaire were also used to calculate daily total energy intake, intake of alcohol, and self-reported vigorous physical activity.
Ascertainment of Incident Lung Cancer Cases and Deaths
Cases were women who were diagnosed during follow-up with incident lung cancer (ICD-9 codes 162.0–162.9), ascertained by means of computerized record linkage to the Canadian Cancer Database. Deaths from all causes were ascertained by means of record linkage to the National Mortality Database. Both databases are maintained by Statistics Canada. The linkages to the databases yielded data on cancer incidence and mortality to 31 December 2000 for women in Ontario, to 31 December 1998 for women in Quebec, and to 31 December 1999 for women in other provinces in Canada. Of the 89,835 women recruited into the study, we excluded 23 lung cancer cases who lacked a date of diagnosis and date of enrollment (7 women) or whose date of diagnosis preceded their date of enrollment (16 women), as well as 24 cases of nonepithelial cancer of the lung. Among the remaining 89,788 women, we identified 750 incident lung cancers. In the subgroup of 49,654 women with dietary information, 42,444 noncases and 342 lung cancer cases had information on weight at age 20.
We used Cox proportional hazards models (using age as the time scale) to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for the association between BMI and lung cancer risk. Study participants were considered at risk from their date of enrollment until the date of diagnosis of their lung cancer, termination of follow-up (the date to which cancer incidence data were available for women in the corresponding province) or death, whichever occurred earliest. Analyses were carried out both on the full cohort (with 750 cases) and on the subgroup for which data on weight at age 20 was available. BMI was categorized into quintiles. Additional models were fitted in the full cohort with BMI categorized into deciles and also as a continuous variable.
Because smoking is the dominant risk factor for lung cancer, all analyses were stratified by smoking status (current, former, and never-smoker). In current smokers, adjustment for number of cigarettes smoked per day, plus years of smoking and pack-years, yielded similar results; we present the data adjusted for pack-years. In former smokers, we adjusted for pack-years plus years since quitting smoking. Additional covariates in the multivariate models were education (3 levels), menopausal status (pre, peri-, postmenopausal), randomization group (intervention versus control), and study center. These models were repeated for subgroups defined by age at diagnosis (dichotomized at the median: <65.5/65.5+), menopausal status (premenopausal/ postmenopausal), and for adenocarcinoma of the lung. Adenocarcinoma was the most common histologic subtype (accounting for 47% of cases with defined histology) followed by small cell carcinoma (16%), squamous cell carcinoma (13%), large cell carcinoma (7%), and mixed or unknown cell types (17%).
Analyses in the subgroup with dietary information additionally included terms for total caloric intake (continuous), vigorous physical activity (any/none), and alcohol consumption (4 levels). Change in weight from age 20 to enrollment was categorized as follows: minimal change (±5 pounds) as the reference group; −6 or more pounds, +6 to 14 pounds, +15 to 24 pounds, +25 to 49 pounds, and +50 pounds or more. To test for trends in the categorical variables of interest, study participants were assigned the median value of their category, and the resulting variable was fitted as a continuous variable in the regression models; the statistical significance of the coefficients was evaluated using the Wald test.31 Use of the Lifetest procedure in SAS (SAS Institute, Cary, NC) showed that the proportional hazards assumption was met. All significance tests were two-sided.
Table 1 presents the baseline characteristics of lung cancer cases and noncases. Cases were older than noncases, had a higher proportion of postmenopausal women, and had higher indices of cigarette smoking. Cases also had lower mean years of education and lower mean body mass index at enrollment. Body mass index at age 20 did not differ between the 2 groups.
After adjusting for age at enrollment, BMI at enrollment differed by smoking status and by case status (Table 2). Among noncases, current smokers as a group had a lower mean BMI compared with both never and former smokers (24.6 current smokers; 25.1 never-smokers; 25.1 former smokers). Among noncases who were current smokers, except for the lightest smoking category, there was a small increase in BMI with increasing number of cigarettes smoked per day. Among lung cancer cases, mean BMI was 26.8 among never-smokers, 24.7 among former smokers, and 24.3 among current smokers. Among cases who were current smokers, BMI was similar over the 4 highest smoking levels; BMI was higher in the lightest smoking category, but this estimate was based on only 15 subjects. At each level of cigarettes per day among current smokers above the lightest smoking level, BMI was lower in cases compared with noncases.
Table 3 presents multivariate-adjusted HRs and 95% CIs for the association of BMI with lung cancer stratified by smoking status. Comparison of the 2 panels shows the effect of adjusting for smoking (pack-years in current smokers and pack-years plus years since quitting in ex-smokers). Among current smokers BMI was inversely associated with risk whether or not pack-years was included in the model, with a slightly stronger trend after inclusion of pack-years (HR for extreme quintiles = 0.63; 95% CI = 0.48−0.83; P for trend = 0.0003). Higher BMI was also associated with reduced risk of lung cancer among former smokers, again with a stronger trend after inclusion of pack-years and years since quitting in the model (HR for extreme quintiles = 0.69; 95% CI = 0.39−1.23; P for trend = 0.02). In contrast, higher BMI was associated with increased risk among never-smokers. The multivariate-adjusted HR for extreme quintiles was 2.19 (95% CI = 1.00−4.80; P for trend 0.007). Similar trends were seen for deciles of BMI and for BMI modeled as a continuous variable (data not shown), although the estimates in former smokers and ex-smokers were unstable due to small numbers. The results in current smokers were similar in analyses stratified on age at diagnosis (<65.5, 65.5+) and menopausal status (premenopausal/postmenopausal). When cases were restricted to those with adenocarcinoma, results were similar.
We repeated the analyses after excluding the 132 lung cancers ascertained within the first 5 years of follow-up. Results were unchanged in all subjects, and in current and former smokers. In never-smokers, however, the HRs were reduced (HR for extreme quartiles = 1.45; 95% CI = 0.64−3.30, and the gradient with higher BMI was weakened (P for trend = 0.11). (The latter analysis was based on only 75 cases). Furthermore, when the follow-up period was partitioned into 3 segments (<8 years, 8–12 years, 13+ years), the inverse association in current smokers was apparent in all 3 time periods. In ex-smokers, the inverse association was apparent particularly in the second and third time periods. In never-smokers, the positive association was seen mainly in the first and third time periods (data not shown).
Among both cases and noncases, weight increased between age 20 and age at enrollment, with a larger increase in noncases (mean 15.9 pounds compared with 14.6 pounds in cases; P < 0.0001). Among current smokers, women who lost more than 5 pounds were not at increased risk compared with women who experienced minimal change in weight (HR = 1.15; 95% CI = 0.72−1.82; Table 4). Among women who gained weight there was a trend toward decreasing risk with increasing level of weight gain (P = 0.004). For women who gained 50+ pounds the HR was 0.50 (95% CI = 0.22−1.12). The effect of change in weight could not be assessed in former and never-smokers due to the small numbers of cases (62 and 45, respectively).
In the present study BMI was inversely associated with lung cancer risk among current and former smokers, whereas in never-smokers it was positively associated with risk. Adjustment for pack-years of smoking (or for duration and intensity of smoking) did not weaken the observed associations in current and former smokers. The inverse association of BMI with lung cancer in current and former smokers was unaffected after excluding lung cancer cases diagnosed in the first 5 years of follow-up, thereby minimizing the likelihood that the observed findings are due to weight loss caused by preclinical disease at enrollment. In addition, among current smokers weight gain between age 20 and age at enrollment was associated with reduced risk.
Our results in smokers are consistent with the findings from several cohort7,11,12 and case-control studies.8,9 Studies that analyzed the association of BMI with lung cancer in never-smokers have shown conflicting results. Three case-control studies8,9,13 and 3 cohort studies7,11,12 provide evidence of an inverse association. In contrast, a large cohort study18 found no association of BMI with lung cancer, and a fourth case-control study19 found a 2.6-fold increased odds ratio for the highest octile of BMI versus the lowest octile in never and long-term former smokers. Our findings in never-smokers are consistent with those of Rauscher et al19; however, our results were attenuated when cases diagnosed within the first 5 years of follow-up were excluded, suggesting that weight gain close to the time of diagnosis in never-smokers may be responsible for the observed positive association.
In smokers, the inverse association of BMI with lung cancer risk could be explained by a number of potential biases. First, if heavier smokers tended to have lower BMI, this could lead to a spurious association. Second, the observed association could be due to preclinical weight loss prior to a diagnosis of lung cancer. Finally, the association could be due to competing mortality from diseases for which obesity is a risk factor. We address each of these points below.
Although current smokers tend to be leaner than former and never-smokers, when attention is restricted to current smokers, those who smoke more tend to have higher BMIs than moderate smokers.21–28 This was true for noncases in our study, although not necessarily for cases. Among noncases who were smokers at baseline, BMI increased modestly over increasing strata of cigarettes smoked per day (Table 2). Among cases, BMI was essentially unchanged over categories of cigarettes per day above the category of 1–10 cigarettes. After adjustment for pack-years of smoking, the association of BMI with lung cancer among current smokers was slightly strengthened rather than attenuated.
Aside from a previous history of breast cancer, no information was available regarding pre-existing illness. However, when lung cancer cases diagnosed during the first 5 years of follow-up were excluded, the association of BMI with lung cancer in current and former smokers was unchanged. In addition, the association of BMI with lung cancer in current smokers was similar when the follow-up period was partitioned into 3 segments. This suggests that the association cannot be explained by preclinical disease prior to baseline. Several other studies have presented similar findings.7,11
Regarding the possibility that competing causes of death could be responsible for the observed association, it should be noted that survival analysis partially takes into account competing causes of death. In the present study the inverse association between BMI and lung cancer was actually stronger in women ages 40–49 at enrollment than in women ages 50–59, suggesting that the association is not an artifact due to early loss of women due to obesity-related conditions.
The paper by Henley et al18 is of particular interest because of the large size of its study population. The authors examined the association of BMI with subsequent lung cancer mortality among 941,105 men and women enrolled in the American Cancer Society's Cancer Prevention Study II and followed for 14 years. To assess bias, they successively eliminated groups in which the association might be confounded. Henley and colleagues concluded that in lifetime nonsmokers who did not report pre-existing disease, leanness was not associated with lung cancer mortality. However, they did not discuss their results in smokers, who accounted for 98% of the lung cancers cases in men and 93% in women. Their data on all cases after exclusion of persons with pre-existing disease and after exclusion of the first 5 years of follow-up tend to support the existence of a modest increased risk associated with low BMI. In men, for those with BMI <20 kg/m2 the hazard ratio was 1.24 (95% CI = 1.05−1.46) after exclusion of those with pre-existing disease and 1.23 (1.01−1.50) after the additional exclusion of the first 5 years of follow-up. The corresponding hazard ratios in women with BMI <19 were 1.32 (1.13−1.55) and 1.28 (1.07−1.53). In women there was suggestive evidence a trend toward increasing risk with decreasing BMI. It is noteworthy that, despite their large study population, Henley et al only categorized number of cigarettes smoked per day as <20, 20, and >20. It is possible that finer adjustment of intensity of smoking would have strengthened the inverse association with BMI.
Few studies have examined weight gain/loss since early adulthood relative to lung cancer risk. In a cohort study from Hawaii, Nomura et al6 found that weight loss since age 25 was associated with increased risk of lung cancer, but found no association between BMI at entry and lung cancer risk. Goodman and Wilkens9 reported in a case-control study that weight gain during adulthood was associated with reduced risk. Both cases and controls tended to gain weight during adulthood, but cases tended to gain less weight, which is very similar to our findings.
Among the strengths of the present study are its prospective nature, the high level of follow-up, and the availability of weight and height measured at enrollment. However, a number of limitations also need to be borne in mind. First, as already mentioned, we did not have information on general health status or history of chronic diseases at enrollment, with the exception of breast cancer and benign breast disease. Second, information on smoking habits was limited to that obtained at baseline. Thus, even though all smoking variables in this study were strong predictors of subsequent lung cancer risk, some misclassification is likely due to changes in smoking habits after enrollment. Furthermore, we did not have information on more detailed aspects of smoking (depth of inhalation, puff frequency, interpuff interval, tar and nicotine content of the cigarette smoked, and mentholation), which may be associated with body mass index and lung cancer risk. Thus, in spite of our adjustment for smoking parameters, there could still be residual confounding of the association of BMI with lung cancer due to smoking.
Two recent studies suggest a possible biologic mechanism underlying the observed inverse association of BMI with lung cancer. In one small study,32 the level of DNA adducts in peripheral lymphocytes was inversely related to BMI level after accounting for the number of cigarettes smoked per day. In another study,33 there was a strong inverse association between BMI and levels of 8-hydroxydeoxyguanosine, a marker of oxidative DNA damage, whereas number of cigarettes smoked per day was not associated with BMI. However, it is unclear from these studies whether leanness itself or some factor correlated with leanness is an independent risk factor for lung cancer, or perhaps modifies the carcinogenic effect of smoking.
In conclusion, the present study contributes to the aggregate evidence suggesting an inverse association between body mass index and lung cancer among smokers. However, the contrasting patterns in ever-smokers and never-smokers, as well as the conflicting results in previous studies of never-smokers, requires explanation. In view of the potential for uncontrolled confounding in most studies to date, the association of body mass index with lung cancer merits further study, particularly in large cohorts with serial measurements of weight and height at regular intervals over the follow-up period, and with updated information on smoking habits and health status.
We thank Statistics Canada, the provincial and territorial Registrars of Vital Statistics, and the Cancer Registry directors for their assistance in making the cancer incidence and mortality data available.
1. Manson JE, Willett WC, Stampfer MJ, et al. Body weight and mortality among women. N Engl J Med
2. Adams KF, Schatzkin A, Harris T, et al. Overweight, obesity, and mortality in a large prospective cohort of persons 50–71 years old. N Engl J Med
3. Lew EA, Garfinkel L. Variations in mortality by weight among 750000 men and women. J Chron Dis
4. Waaler HT. Height, weight and mortality: the Norwegian experience. Acta Med Scand
. 1984;(Suppl 679):1–56.
5. Garn SM, Hawthorne VM, Pilkington JJ, et al. Fatness and mortality in the West of Scotland. Am J Clin Nutr
6. Nomura AMY, Heilbrun LK, Stemmermann GN. Body mass index as a predictor of cancer in men. J Natl Cancer Inst
7. Knekt P, Heliovaara M, Rissanen A, et al. Leanness and lung cancer. Int J Cancer
8. Kabat GC, Wynder EL. Body mass index and lung cancer risk. Am J Epidemiol
9. Goodman MT, Wilkens LR. Relation of body size and the risk of lung cancer. Nutr Cancer
10. Chyou P-H, Nomura AMY, Stemmermann GN. A prospective study of weight, body mass index and other anthropometric measurements in relation to site-specific cancer. Int J Cancer
11. Kark JD, Yaari S, Rasooly I, et al. Are lean smokers at increased risk of lung cancer? Arch Intern Med
12. Olson JE, Yang P, Schmitz I, et al. Differential association of body mass index and fat distribution with three major histologic types of lung cancer: evidence from a cohort of older women. Am J Epidemiol
13. Kubik AK, Zatloukal P, Tomasek L, et al. Dietary habits and lung cancer risk among non-smoking women. Europ J Cancer Prev
14. Kanashiki M, Sairenchi T, Saito Y, et al. Body mass index and lung cancer: a case-control study of subjects participating in a mass-screening program. Chest
15. Lee I-M, Paffenbarger PS. Re: Body mass index and lung cancer risk [letter]. Am J Epidemiol
16. Eichholzer M, Bernasconi F, Jordan P, et al. Body mass index and the risk of male cancer mortality of various sites: 17-year follow-up of the Basel cohort study. Swiss Med Wkly
17. Drinkard CR, Sellers TA, Potter JD, et al. Association of body mass index and body fat distribution with lung cancer in older women. Am J Epidemiol
18. Henley SJ, Flanders DW, Manatunga A, et al. Leanness and lung cancer risk: fact or artifact? Epidemiology
19. Rauscher GH, Mayne ST, Janerich DT. Relation between body mass index and lung cancer risk in men and women never and former smokers. Am J Epidemiol
20. Albanes D. Caloric intake, body weight, and cancer: a review. Nutr Cancer
21. Noppa H, Bengtsson C. Obesity in relation to smoking: a population study of women in Goteborg, Sweden. Prev Med
22. Jacobs DR, Gottenberg S. Smoking and weight: the Minnesota Lipid Research Clinic. Am J Public Health
23. Albanes D, Jones DY, Micozzi MS, et al. Associations between smoking and body weight in the US population: analysis of NHANES II. Am J Public Health
24. Fehily AM, Phillips KM, Yarnell JW. Diet, smoking, social class, and body mass index in the Caerphilly Heart Disease Study. Am J Clin Nutr
25. Seidell JC, Cigolini M, Deslypere JP, et al. Body fat distribution in relation to physical activity and smoking habits in 38-year-old European men. The European Fat Distribution Study. Am J Epidemiol
26. Shimokata H, Muller DC, Andres R. Studies in the distribution of body fat. III. Effects of cigarette smoking. JAMA
27. Marti B, Tuomilehto J, Korhonen HJ, et al. Smoking and leanness: evidence for change in Finland. BMJ
28. Kabat GC, Wynder EL. Re: Body mass index and lung cancer risk [letter]. Am J Epidemiol
29. Miller AB, Baines CJ, To T, et al. Canadian National Breast Screening Study: 1. Breast cancer detection and death rates among women aged 40 to 49 years. CMAJ
30. Miller AB, Baines CJ, To T, et al. Canadian National Breast Screening Study: 2. Breast cancer detection and death rates among women aged 50 to 59 years. CMAJ
31. Rothman KJ, Greenland S. Modern Epidemiology
, 2nd ed. Philadelphia: Lippincott-Raven; 1998.
32. Godschalk RWL, Feldker DEM, Borm PJA, et al. Body mass index modulates aromatic DNA adduct levels and their persistence in smokers. Cancer Epidemiol Biomarkers Prev
33. Mizoue T, Kasai H, Kubo T, et al. Leanness, smoking, and enhanced oxidative DNA damage. Cancer Epidemiol Biomarkers Prev