The association between change in body mass index and upper aerodigestive tract cancers in the ARCAGE project: Multicenter case–control study
Article first published online: 20 MAY 2010
Copyright © 2010 UICC
International Journal of Cancer
Volume 128, Issue 6, pages 1449–1461, 15 March 2011
How to Cite
Park, S. L., Lee, Y.-C. A., Marron, M., Agudo, A., Ahrens, W., Barzan, L., Bencko, V., Benhamou, S., Bouchardy, C., Canova, C., Castellsague, X., Conway, D. I., Healy, C. M., Holcátová, I., Kjaerheim, K., Lagiou, P., Lowry, R. J., Macfarlane, T. V., Macfarlane, G. J., McCartan, B. E., McKinney, P. A., Merletti, F., Pohlabeln, H., Richiardi, L., Simonato, L., Sneddon, L., Talamini, R., Trichopoulos, D., Znaor, A., Brennan, P. and Hashibe, M. (2011), The association between change in body mass index and upper aerodigestive tract cancers in the ARCAGE project: Multicenter case–control study. Int. J. Cancer, 128: 1449–1461. doi: 10.1002/ijc.25468
- Issue published online: 28 JAN 2011
- Article first published online: 20 MAY 2010
- Manuscript Accepted: 22 APR 2010
- Manuscript Received: 28 OCT 2009
- European Commission (5th Framework Programme). Grant Number: QLK1-CT-2001-00182
- International Agency for Research on Cancer (Special Training Award)
- Italian Association for Cancer Research
- Compagnia di San Paolo/FIRMS
- National Institute of Health/National Cancer Institute. Grant Number: T32 CA09142
- BMI change;
- upper aerodigestive tract cancers
Previous studies reported an inverse relationship between body mass index (BMI) and upper aerodigestive tract (UADT) cancers. Examining change in BMI over time may clarify these previous observations. We used data from 2,048 cases and 2,173 hospital- and population-based controls from ten European countries (alcohol-related cancers and genetic susceptibility in Europe study) to investigate the relationship with BMI and adult change in BMI on UADT cancer risk. Odds ratios (ORs) and 95% confidence intervals (CIs) were estimated for associations between BMI at three time intervals and BMI change on UADT cancer development, adjusting for center, age, sex, education, fruit and vegetable intake, smoking and alcohol consumption. We found an inverse relationship between UADT cancers and BMI at time of interview and 2 years before interview. No association was found with BMI at 30 years of age. Regarding BMI change between age 30 and 2 years before interview, BMI decrease (BMI change <−5%) vs. BMI stability (−5% ≤ BMI change <5%) showed no overall association with UADT cancers (OR = 1.15; 95% CI = 0.89, 1.49). An increase in BMI (BMI change ≥+5%) was inversely associated with UADT cancers (OR = 0.74; 95% CI = 0.62, 0.89). BMI gain remained inversely associated across all subsites except for esophageal cancer. When stratified by smoking or by drinking, association with BMI gain was detected only in drinkers and smokers. In conclusion, BMI gain is inversely associated with UADT cancers. These findings may be influenced by smoking and/or drinking behaviors and/or the development of preclinical UADT cancers and should be corroborated in studies of a prospective nature.
Cancers of the upper aerodigestive tract (UADT) have been primarily attributed to tobacco smoking and alcohol consumption.1, 2 However, at least 25% of cases are potentially due to other factors, such as human papillomavirus, poor diet, low socioeconomic status (SES) and potentially lean body size.3–7 Contrary to many other cancer sites such as the breast and colon, where obesity is positively associated with cancer development,8 previous studies have observed an inverse relationship between body mass index (BMI) and squamous cell carcinoma (SCC) UADT cancers (where leanness is positively associated and higher BMI is inversely associated).5, 9–24
The potential biological mechanism or explanation for these observations remains unclear. These findings, the majority of which are from case–control studies, may be a result of reverse causation, where cancer status affects BMI, and/or residual confounding. To discern this relationship, studies have examined BMI at various time points while adjusting for confounding variables. Studies found lean BMI few years before diagnosis positively12-14, 24 and not associated25 with UADT or head-and-neck cancers, respectively. BMI at young adulthood (20–30 years old) has been found to be inversely9, 11, 25 and not associated12, 14, 24 with these cancer sites. Among prospective studies that adjusted for smoking, this relationship was evaluated in esophageal squamous cell carcinoma (ESCC)16, 17, 19, 20, 23 and laryngeal cancer.19
These prior results do not explain whether the observed effect is due to leanness as a result of change in BMI over time, particularly because smoking and alcohol drinking can affect body weight.26–30 Thus, investigating the role of BMI change in UADT cancers may be an effective means to better understand the prior findings. To our knowledge, only one study investigated this potential relationship with weight change in ESCC.11 Here, we will investigate the association between BMI change and UADT cancers between two time points (2 years before interview and at 30 years of age), using data from the alcohol-related cancers and genetic susceptibility in Europe (ARCAGE) study.
Material and Methods
Alcohol-related cancers and genetic susceptibility in Europe study
Details regarding the ARCAGE study have been previously published.31 In short, our study was designed to investigate and clarify the role of smoking, drinking and genetic factors in UADT cancers. ARCAGE was initiated by the International Agency for Research on Cancer in 2002, with the participation of 14 centers in ten European countries. Recruitment was conducted from 2002 to 2005 for all centers, except for the French center, where recruitment was conducted during 1987–1992. Cases were identified by participating hospitals within 6 months of diagnosis. Eligibility was determined using ICD-O codes: C00, C01, C02, C03, C04, C05, C06, C09, C10, C12, C13, C14.0, C14.8, C15.0, C15.3, C15.4, C15.5, C15.8, C15.9 and C32. Controls were frequency matched to cases by sex, age (in 5-year intervals) and referral (or residence) area. ARCAGE centers used hospital controls, except for centers in the United Kingdom (UK), which used population-based controls. Hospital controls were randomly selected from subjects admitted as in- or outpatients in the same hospital as the case. Eligibility of controls included recent disease diagnosis, short hospital stay (the majority ≤ 1 week) and did not have admission diagnoses related to alcohol, tobacco or dietary practices. Eligible control admission diagnosis included (i) endocrine and metabolic, (ii) genitor-urinary, (iii) skin, subcutaneous tissue and musculoskeletal, (iv) gastrointestinal, (v) circulatory, (vi) ear, eye and mastoid and (vii) nervous system diseases, as well as (viii) plastic surgery cases and (ix) trauma patients. The proportion of controls within a specific diagnostic group did not exceed 33% of the total. In the UK centers, population-based controls were recruited from a randomly selected list of ten controls for every case, matched by age, sex and same family medical practice.
Epidemiological data collection was performed by trained interviewers using a lifestyle questionnaire consisting of questions regarding sociodemographic factors, detailed smoking history, environmental tobacco smoke exposure for nonsmokers, history of alcohol drinking, personal medical history of diseases associated with UADT cancers, oral cavity health, lifetime occupational history and dietary habits based on a semiquantitative food frequency questionnaire. Detailed smoking and alcohol history includes age and frequency of use when this habit was initiated, altered or ceased, if applicable. Variables for age at initiation and cessation, duration, frequency and time since cessation are available. Alcohol drinking was defined by average alcohol drinking frequency 1 year before interview. Data were collected from a total of 2,286 cases (2,109 SCC and 151 non-SCC cases and 26 with missing histological information) and 2,227 controls. UADT cancers included the following subsites: oral, oropharyngeal, hypopharyngeal, laryngeal and esophageal cancers. For the primary analysis, we excluded non-SCC cases because of limitation in sample size. Of the total 2,109 SCC cases, we excluded 61 cases and 54 controls because of missing at least one of the following adjustment variables: sex, age, smoking status, pack-years smoked, alcohol frequency, education or fruit and vegetable consumption frequency. Adenocarcinomas of the esophagus were evaluated separately.
Both self-reported height and weight at interview were recorded by trained interviewers, except for UK and Dublin centers, where weight at interview was not recorded. Weight measures 2 years before interview and at age 30 were also self-reported at time of interview, but were not reported in the Paris center. BMI (kg/m2) and percent BMI change were calculated from these height and weight measures. Thus, BMI change at 2 years before interview was not available for all three centers, and BMI change from age 30 to 2 years prior was not available for the Paris center. We used cut points for BMI categories as defined by the World Health Organization standards.32 BMI change, our primary variable of interest, is defined as a percentage [(BMI at 2 years before interview − BMI at age 30)/BMI at age 30]×100. We selected this measure because it standardizes BMI differences to ones' referent BMI (at age 30), as well as potentially accounting for any associations related to height. To determine definition of BMI loss, gain and stability, we explored associations between UADT cancers and percent changes in finer categories (2.5%). In preliminary findings, we found that “stable” appeared to be within the 10% interval of −5% ≤ BMI change <5%. Given there was no dose response within levels of gain and levels of loss and preliminary findings support collapsibility, the final collapsed BMI change categories were as follows: loss (BMI change <−5%), stable (−5% ≤ BMI change <5%) and gain (BMI change ≥+5%).
All statistical analyses were performed using SAS v. 9.1.3 (Cary, NC). Differences between baseline characteristic distributions between cases and controls were evaluated using χ2 tests; differences in BMI at 2 years before interview and BMI change by baseline characteristics amongst cases and controls were evaluated using t-test for comparisons between two categories (i.e., gender) and one-way ANOVA for variables with >2 categories. Heterogeneity by center was examined using the likelihood ratio test, which tested the difference between the log likelihood of the model with the product term, center and BMI change, and the model without the product term, based on a χ2 distribution.
Unconditional logistic regression models were used to estimate the odds ratios (OR) and 95% confidence intervals (CI). The models were adjusted for the following variables: center, sex, age (in 5-year intervals), smoking status including pack-years (seven categories: never smoker; former smoker: 0 < pack-years ≤ 20, 20 < pack-years ≤ 40 or pack-years >40 and current smoker: 0 < pack-years ≤ 20, 20 < pack-years ≤ 40 or pack-years >40), fruit and vegetable intake (low, medium and high), education (finished primary school, finished further school and university degree) and alcohol consumption frequency (never, <1 drink per day, 1–2 drinks/day, 3–4 drinks/day and 5+drinks/day).
To increase precision and to preserve the previously observed dose-response association between smoking and UADT cancers,33 we selected the aforementioned smoking adjustment variable, instead of adding to the model smoking status and pack-years as two separate variables. Height was categorized in gender-specific quintiles among controls. The “fruit and vegetable” adjustment variable was created by combining center-specific fruit and vegetable consumption tertiles, which were previously determined.34 We defined “low fruit and vegetable intake” as having both center-specific fruit and vegetable intakes in the lowest tertile or at least one group in the lowest tertile and the other group in the mid-tertile; “mid fruit and vegetable intake” incorporates either both center-specific fruit and vegetable in the middle tertile or having one of each fruit and vegetable intake at either extreme tertile (i.e., lowest tertile in fruit and highest tertile in vegetable) and “high fruit and vegetable intake” as having both intakes in the highest tertile or one in the high and one in the middle tertile. There was no difference in our findings when including separate fruit and vegetable tertile variables or the combined variable into the model. The combined variable was selected to maximize precision. Tests for linear trend were conducted by including the variable of interest as a continuous variable. Tests for interactions were assessed by comparing the fit of the full regression model to the full regression model including interaction terms. Interaction terms were the product terms of the categories as described above and the categorical BMI change variable.
Table 1 presents the baseline characteristics of 2,048 SCC cases and 2,173 controls, after the exclusion of cases and controls with missing adjustment variable information. Cases were more likely to be a male smoker, drinker, have a lower education and lower fruit and vegetable intake compared to controls. Body size distribution differed by age and tobacco use in both cases and controls, by education in controls and by alcohol use and fruit and vegetable intake in cases. The distribution of BMI change, between 30 years of age and 2 years before interview, in both cases and controls varied by center, sex, age and tobacco use. We also found a difference in mean BMI change between former and current smokers, where former smokers showed a greater BMI increase (p < 0.001). Among cases, distribution of BMI change differed by frequency of alcohol and fruit/vegetable intake. We evaluated correlations between our variables in our control population and found that there was little correlation between smoking and drinking status, current BMI and drinking or smoking status (r = 0.17, r = 0.11 and r = −0.13, respectively).
Associations between BMI at various time points and BMI change and UADT cancers can be found in Table 2. Leanness (13.0–18.4 kg/m2) at time of interview and 2 years prior was associated with UADT cancers (OR = 1.90; 95% CI = 1.28, 2.82 and OR = 2.10; 95% CI = 1.16, 3.81, respectively), whereas heavier BMIs (25.0–29.9 kg/m2 or 30.0–53.0 kg/m2) at 2 years before interview were inversely associated with UADT cancers (OR = 0.74; 95% CI = 0.62, 0.88 and OR = 0.74; 95% CI = 0.59, 0.93, respectively). Results for men and women were consistent (data not shown). There was no association between UADT cancers and BMI at age 30. BMI loss during the 2 years before interview showed a positive association with UADT cancers. BMI gain from age 30 to 2 years before interview was inversely related with UADT cancer development (OR = 0.74; 95% CI = 0.62, 0.89). When including BMI at 2 years into the model, the association with BMI gain in UADT cancers was only slightly attenuated (OR = 0.83; 95% CI = 0.69, 1.02, data not shown).
Results for BMI measures did not vary when stratified by subsite (Table 3). BMI loss at 2 years before interview was positively associated with all UADT cancer subsites. There was a suggestive positive association between BMI loss at age 30 to 2 years before interview and oral/oropharyngeal cancers (OR = 1.22; 95% CI = 0.90, 1.64). BMI gain remained inversely associated across all cancer subsites except for the esophagus (both SCC and adenocarcinoma), where we found no association, possibly a result of small sample size. Among adenocarcinoma of the esophagus, we observed a positive association in BMI ≥30.0 kg/m2 at 2 years before study entry (OR = 2.89; 95% CI = 1.18, 7.06, results not presented). Test for heterogeneity among centers was p = 0.059. When stratified by centers, findings were consistent, with the exception of Barcelona, where BMI loss was inversely related to UADT cancers (OR = 0.29; 95% CI = 0.11, 0.71, data not shown). When selectively removing each individual center from the analyses, Barcelona appeared to have the greatest influence; upon removal, BMI loss was positively associated with UADT cancers (OR = 1.34; 95% CI = 1.02, 1.75, data not shown).
Table 4 presents effect modification by UADT cancer risk factors: tobacco smoking, alcohol drinking, vegetable and fruit intake, BMI at age 30 and time to disease. No association was observed between BMI change among never or former smokers; however, among current smokers, BMI gain was inversely associated with UADT cancers (OR = 0.59; 95% CI = 0.46, 0.76). Both former alcohol drinkers and current drinkers also showed inverse associations between BMI gain and UADT cancers, and in current drinkers, a suggestive positive association was observed with BMI loss (OR = 1.28; 95% CI = 0.94, 1.74). Among those with low fruit and vegetable consumption, BMI gain was inversely associated with UADT tumorigenesis, whereas, in those with higher fruit and vegetable consumption, BMI loss was positively associated. BMI gain was inversely associated with UADT cancers among those with BMIs ranging from 18.5 to 30.0 kg/m2 at age 30; however, at heavier BMIs, BMI gain or loss did not appear to influence the development of UADT cancers. Associations with BMI gain from age 30 to 2 years before interview diminished with increasing time to disease. Interactions were not detected with smoking, alcohol drinking and fruit and vegetable intake (p values = 0.071, 0.169 and 0.128, respectively) and were detected with BMI at age 30 and time to disease (p values = 0.003 and 0.034, respectively). When testing heterogeneity between strata, a difference in ORs was present for BMI gain and UADT cancers between former and current smokers (p = 0.03); this difference was not notable when comparing ORs between never and current smokers (p = 0.09).
In our study population from ten European countries, similar to prior studies, we found an inverse association between UADT cancers and BMI at time of interview and 2 years before interview.5, 9–24 After adjusting for confounding variables, we found BMI loss 2 years before interview associated with UADT cancers, whereas BMI gain during the period of 30 years to 2 years before interview was inversely associated with UADT cancers. Stratified analyses suggest that this inverse relationship may be influenced by smoking habits and accuracy of report.
We believe that the association between leanness at diagnosis is likely a result of inverse causality. The observation that BMI loss 2 years before diagnosis is associated with UADT cancers verifies this. Regarding earlier BMI measures, it has been suggested that leanness may be an earlier marker for cancer development of the oral cavity and pharynx9 or due to residual confounding from the relationship between weight, smoking and drinking.26–29 Smokers are more likely to drink alcohol35 and often have a lower BMI than nonsmokers; however, heavier smokers tend to be of slightly greater BMI than lighter smokers.27 Moderate alcohol drinking has also been associated with a reduction in body weight.29, 36 However, among controls in our dataset, correlations were weak between smoking and drinking status and between current BMI and drinking or smoking status.
Prior case–control studies found inverse associations in measures as early as age 30 (men only)9 and 5 years before diagnosis.15 In an international pooling of case–control studies, Gaudet et al. found a nonsignificant positive association in lean BMIs 2 to 5 years before diagnoses (OR = 1.56; 95% CI: 0.80, 3.02).25 Findings from cohort studies may elucidate these associations, because they are not subject to recall bias, and the nature of the study design decreases the possibility of reverse causality. In a cohort study that had the longest follow-up time of 10 years with adjustment for both smoking and alcohol drinking, the investigators found that heavier BMI appears “protective” against ESCC.20 In addition, a meta-analysis showed that case–controls studies report a greater inverse relative risk of ESCC per 5 kg/m2 BMI increase (OR = 0.49; 95% CI = 0.44, 0.55) than pooled estimates of cohort studies (RR = 0.69; 95% CI = 0.63, 0.75).20 Cohort studies on oral, pharyngeal and laryngeal cancers that adjusted for smoking and alcohol drinking are not available thus far.
Although BMI at age 30 and BMI loss is not associated with UADT cancers, we observed BMI gain from age 30 to 2 years before interview to be inversely associated. We are aware of only one study investigating the association between weight change and the UADT subsite, ESCC. Similar to our study, Gallus et al. found that weight increases were inversely associated; however, unlike our findings, decrease in weight also had an inverse relationship with ESCC.11 Differences are possibly a result of smaller sample size in this European study by Gallus et al. (weight change cases = 293). Interestingly, BMI gain is not associated with UADT cancers in the heaviest BMI categories (≥30.0 kg/m2), possibly as a result of obesity-related competing risks, because mortality substantially increases with greater obesity,37 which should be corroborated in a prospective study.
BMI gain may be inversely associated with UADT cancer development due to residual confounding by smoking and/or alcohol drinking. When additional smoking duration and drinking consumption measures were included in the regression model, findings remained consistent. On the other hand, if history of smoking or alcohol drinking was differentially misreported because of case status or frequency of usage, our ability to correct for residual confounding may be limited. We do believe, however, that our smoking and drinking measures must have some degree of accuracy as our data showed similar findings as previously conducted studies regarding the relationship of smoking and drinking status and dose response of smoking and drinking with UADT cancer risk.
We considered the possibility that the inverse relationship between BMI gain and UADT cancers may be related to smoking cessation as former smokers often have heavier BMIs than current smokers.30 In our control population, we found former smokers had greater BMI gain than current smokers. Former smokers may gain weight after quitting,30 and their risk of UADT cancers will be decreasing with increasing years of cessation.38 However, an association between BMI gain and UADT cancers among former smokers was not found. In addition, the inverse association between BMI gain and UADT cancers was present only among former smokers who recently quit smoking (>1–4 years). In current smokers, no difference in mean BMI change was present among those who claimed to have intermittently ceased or decreased their smoking habits. Thus, it is unlikely that our observed results are due to smoking cessation.
In stratified analyses, the inverse associations between BMI gain and UADT cancers were present exclusively in current smokers and former or current drinkers. We also examined this association by duration of smoking and drinking and found that the inverse association is present primarily among those with longer duration of drinking (≥20 years) or smoking (≥30 years). In support of our findings, many studies have found an inverse associations between heavier BMIs and UADT cancers only among those with a history of smoking,5, 9, 11, 20, 21, 24 although tests for effect modification have shown conflicting results. We did not detect an interaction between smoking or drinking and BMI change. For smoking status, the lack of evidence of interaction might be due to the smaller sample size in the never smokers. In another study, the investigators found an interaction between waist–hip ratio (WHR) and alcohol drinking in ESCC risk, but not with BMI,21 suggesting that we did not detect an interaction between alcohol drinking and BMI change because BMI, which is not perfectly correlated with adult body fat distributions,39 was unable to capture the “true” interaction that exists between WHR and alcohol drinking.
Our findings may be associated with hormonal or growth factor pathways. It has been observed that estrogen is inversely associated with prevalence of oral leukoplakia40; thus, it may be possible that in women higher estrogen levels as a result of increasing BMI41 could counterbalance the antiestrogenic effects seen in smoking.42 Additional understanding of the role of estrogen in UADT cancer as well as adjustment for hormone replacement therapy and menopausal status is necessary. Among men, studies have found an inverse association between insulin-like growth factor-1 level and smoking,43–45 suggesting that male smokers with weight gain may have lower levels of this mitogenic and antiapoptotic protein,46 thereby reducing the risk of developing UADT cancers. Further understanding of these pathways and their interaction with smoking and alcohol are needed. Lastly, given that the general adult population gains weight with age, among those who are unable to increase their adult BMI, our findings may indicate a presence of early-stage disease in smokers and drinkers. A study found that within a 2-year period small levels of adult weight gain (1 to <2.9 BMI units) did not change mortality risk compared to weight loss and excessive weight gain, which increased mortality.37
There were some limitations to our study. Selection bias from use of hospital-based controls may be present; however, when we restricted our analysis to only population-based controls, findings were consistent. It is possible that some diagnoses of hospital-based controls (e.g., metabolic or digestive) may be related to BMI measures. However, we found no notable difference in mean BMI change from age 30 to 2 years before interview among hospital controls with and without these disorders. In addition, when excluding these controls from our analysis, results were similar. Potential heterogeneity was detected between centers, which may be attributed to the Barcelona center, where an inverse relationship was found with both BMI loss and gain. No explanation could be identified and warrants further investigation. It may be chance finding or due to factors that influence BMI change, such as stage of disease. Additionally, use of a semiquantitative food frequency questionnaire limited the possibility to account for energy balance in our BMI change models.
Self-reported weight measures at 2 years before interview and at age 30 may have differential misclassification, due to perceived weight, or difference in reporting by age or hospital-based controls. However, we found that mean BMI change in hospital- and population-based controls did not differ. The associations between BMI change and UADT cancers attenuated toward the null in the older subjects, which might be due to a decrease in accuracy of recalled weight. Self-reported weight measures recalled within 20 years can be fairly accurate47; however, with older age, underestimation of reported BMI measures increases.48, 49 This inaccuracy, however, has correlation values (r) > 0.69.48, 49 We were unable to report on possible weight fluctuations between our BMI change measures, adult height loss and weight measures for all study centers. It would be of interest to evaluate such measures in future studies.
Strengths of the study include large sample size, BMI measures at different time points and adjustment for smoking and drinking measures. In conclusion, BMI gain is inversely associated with UADT cancers. Explanation for this finding remains unclear; however, we could speculate that this may be due to residual confounding, interactions of body fat distribution with smoking and/or drinking, biological mechanisms or indication of early tumor development. To further investigate these hypotheses, our results should be corroborated in studies of a prospective nature.
The authors gratefully acknowledge the study interviewers and their clinical colleagues in hospitals and primary care who supported this study, and especially, they thank the study participants and their families for their participation. Dr. Yuan-Chin Amy Lee, Ms. Sungshim Lani Park and Ms. Manuela Marron were sponsored by a special training award at the International Agency for Research on Cancer.
- 32World Health Organization. Obesity: preventing and managing the global epidemiced. Geneva: WHO, 2000.
- 44Serum insulin-like growth factor I in a random population sample of men and women: relation to age, sex, smoking habits, coffee consumption and physical activity, blood pressure and concentrations of plasma lipids, fibrinogen, parathyroid hormone and osteocalcin. Clin Endocrinol (Oxf) 1994; 41: 351–7., , , , , , .