Skip Navigation


American Journal of Epidemiology Advance Access originally published online on May 15, 2008
American Journal of Epidemiology 2008 168(1):105-114; doi:10.1093/aje/kwn091
This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow All Versions of this Article:
168/1/105    most recent
kwn091v1
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrow Search for citing articles in:
ISI Web of Science (8)
Right arrowRequest Permissions
Right arrow Disclaimer
Google Scholar
Right arrow Articles by Pandeya, N.
Right arrow Articles by Whiteman, D. C.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Pandeya, N.
Right arrow Articles by Whiteman, D. C.
Social Bookmarking
Add to CiteULike   Add to Connotea   Add to Del.icio.us  
What's this?

American Journal of Epidemiology © The Author 2008. Published by the Johns Hopkins Bloomberg School of Public Health. All rights reserved. For permissions, please e-mail: journals.permissions@oxfordjournals.org.

PRACTICE OF EPIDEMIOLOGY

Associations of Duration, Intensity, and Quantity of Smoking with Adenocarcinoma and Squamous Cell Carcinoma of the Esophagus

Nirmala Pandeya1,2, Gail M. Williams2, Shahram Sadhegi1,2, Adèle C. Green1, Penelope M. Webb1 and David C. Whiteman1

1 Population Studies and Human Genetics Division, Queensland Institute of Medical Research, Herston, Queensland, Australia
2 School of Population Health, University of Queensland, Brisbane, Queensland, Australia

Correspondence to Dr. David Whiteman, Queensland Institute of Medical Research, PO Royal Brisbane Hospital, Brisbane, Queensland 4029, Australia (e-mail: david.whiteman{at}qimr.edu.au)

Received for publication December 12, 2007. Accepted for publication March 18, 2008.


   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 References
 
Smoking has been identified as a risk factor for esophageal cancer; however, there is evidence that magnitudes and patterns of association differ by histologic type. The authors aimed to measure and compare the independent effects of various dimensions of smoking (duration, intensity, total dose, and time since quitting) on risks of esophageal adenocarcinoma (EAC), gastroesophageal junction adenocarcinoma (GEJAC), and esophageal squamous cell carcinoma (ESCC). They used data from a population-based Australian case-control study (2002–2005) comprising 367 EAC cases, 426 GEJAC cases, and 309 ESCC cases and 1,580 controls. Multivariate logistic and generalized additive logistic regression (for nonlinear dose effects) were used. Ever smokers had significantly higher risks of EAC (odds ratio (OR) = 1.7, 95% confidence interval (CI): 1.3, 2.3), GEJAC (OR = 2.4, 95% CI: 1.8, 3.2), and ESCC (OR = 2.8, 95% CI: 2.0, 4.0) than did never smokers; however, there were significant differences in magnitude and patterns of association between subtypes. When multiple dimensions of smoking were assessed concurrently, duration was significantly associated with all three subtypes but intensity was associated only with GEJAC and ESCC, and the associations were nonlinear. For all types of esophageal cancer, time since quitting was independently associated with approximately 15–19% risk reductions per decade.

esophageal neoplasms; smoking

Abbreviations: CI, confidence interval; EAC, esophageal adenocarcinoma; ESCC, esophageal squamous cell carcinoma; GEJAC, gastroesophageal junction adenocarcinoma; OR, odds ratio


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 References
 
Tobacco smoking is an established risk factor for both of the common histologic types of esophageal cancer; however, there is evidence that the patterns of risk differ for each (1–3). Thus, while most epidemiologic studies of esophageal squamous cell carcinoma (ESCC) have reported monotonic risk increases with increasing levels of smoking (1–5), studies of esophageal adenocarcinoma (EAC) and gastroesophageal junction adenocarcinoma (GEJAC) have typically reported associations of smaller magnitude (3, 6–8). Such findings suggest that smoking may cause different types of esophageal cancer through different mechanisms.

Past exposure to tobacco smoke is typically measured across dimensions such as duration, intensity, cumulative dose, and time since quitting (among ex-smokers). Each dimension can be assessed independently as a risk factor for esophageal cancer. Studies investigating the role of smoking in cancer at other sites (notably the lung) have shown that different dimensions of smoking may have independent or joint associations with cancer, depending upon the histologic subtype(s) under consideration. In addition to the potential for deriving mechanistic insights, assessing the role of the various components of smoking exposure in cancer risk is necessary for estimating the likely impact of smoking control strategies and optimizing evidence-based health advice. While investigators have previously reported on the risks of esophageal cancer associated with smoking (2, 3, 8–10), none (to our knowledge) have assessed the independent contribution of each dimension of smoking to determine which, if any, is of primary importance.

We analyzed the effects of multiple dimensions of smoking exposure on esophageal cancer risk using approaches similar to those developed to investigate the causes of lung cancer (4, 11, 12). Specifically, we sought to quantify the independent associations between dimensions of smoking and their dose patterns and the risks of esophageal and gastroesophageal cancer and to explore possible subtype-specific differences in these associations.


   MATERIALS AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
 References
 
Study design and participants
We conducted a population based case-control study of esophageal cancer. Full details on the study design and recruitment have been published previously (13). Briefly, eligible case patients were people aged 18–79 years with a histologically confirmed primary invasive cancer of the esophagus or gastroesophageal junction diagnosed between July 1, 2002 (July 1, 2001, in Queensland) and June 30, 2005, in mainland states of Australia. Patients were recruited through either major treatment centers or state-based cancer registries. Through treatment centers, we identified 1,610 eligible patients; of these, 167 died before consent could be obtained, the treating doctor denied permission to contact 71, and 181 were excluded (91 were too ill, 41 were unable to communicate in English, 23 were mentally incapable, and 26 could not be contacted). Through the cancer registries, of 739 eligible patients who were alive at the time of identification, the treating doctor denied permission to contact 84, 37 were either physically or mentally incapable of participating, and 232 could not be contacted. (A further 835 persons with registry notification of "esophageal cancer" had died before study identification, and their records could not be verified.) Of the 1,577 patients with esophageal cancer invited to participate in the study (1,191 through clinics and 386 through cancer registries), 1,102 patients returned a completed questionnaire (70 percent of all patients invited; 35 percent of all living and deceased persons in mainland Australia who had been diagnosed with incident esophageal cancer). Final numbers of case participants, by histologic type, were: EAC, 367; GEJAC, 426; and ESCC, 309.

Potential controls were randomly selected from the Australian electoral roll by 5-year age group and state of residence to match the distribution of the case series. We aimed for similar numbers of male cases and controls in each stratum of age and state; female controls were intentionally oversampled at all ages to accommodate their simultaneous enrollment in a case-control study of ovarian cancer (14). Of 3,258 potentially eligible control participants who were invited to participate, 175 were excluded because they were deceased (n = 16), too ill (n = 61), or unable to communicate in English (n = 98), and 41 were lost to follow-up between initial contact and participation. Of the 3,042 remaining controls, 1,680 (55 percent) accepted. Completed questionnaires were returned by 1,580 controls (49 percent of all potentially eligible controls).

Data collection
Health and lifestyle factors.
Participants self-completed a health and lifestyle questionnaire asking about their education, height, and weight 1 year prior (1 year before diagnosis for cases). We calculated body mass index by dividing weight (kg) by height (m) squared and used standard categories for analysis (healthy weight: <25.0; overweight: 25.0–29.9; obese: ≥30). Participants were also asked about their past alcohol consumption, frequency of heartburn ("a burning pain behind the breastbone after eating") or acid reflux ("a sour taste from acid or bile rising up into the mouth or throat"), and aspirin use in the past 5 years.

Smoking history.
Participants were asked whether, over their whole life, they had ever smoked more than 100 cigarettes or cigars (or equivalent use of pipes). Positive responses led to further questions about the age at which they started smoking, the age at which they stopped smoking permanently (among ex-smokers), the amount they smoked in a typical day, the number of days per week that they smoked, and the number of years they had smoked. We asked participants to report the duration of each episode of temporarily stopping smoking for more than 6 months and their age at the time. In addition, participants were asked to estimate the average number of cigarettes, cigars, or pipes smoked per day during each decade of life.

Derivation of smoking variables.
Current smokers and ex-smokers were defined by their smoking status at 1 year prior to their reference age (age at diagnosis for cases, age at study participation for controls). Smoking duration was defined as the difference between starting age and either permanent quitting age (ex-smokers) or reference age (current smokers), after subtracting the cumulative duration of any episodes of temporarily quitting smoking. Smoking intensity was defined as the average number of cigarettes smoked in a typical day.

We estimated each participant's lifetime cumulative quantity of tobacco smoked (dose) in pack-years, using two algorithms. We used a simple algorithm to calculate the product of smoking duration and intensity. We used a detailed algorithm to calculate the sum of decade-specific smoking doses, where each decade-specific smoking dose was the product of the smoking intensity during that decade, the number of days of smoking per week, and smoking duration within that decade. The correlation between doses calculated by means of the two algorithms was very high (Pearson correlation = 0.95); however, the detailed algorithm resulted in significantly lower dose estimates (~2 pack-years). For these analyses, we used the cumulative smoking dose calculated by means of the detailed algorithm.

Time since quitting was calculated as the difference between the age at which ex-smokers had permanently stopped smoking and their age 1 year prior to their reference age. For categorical data analysis, each of the continuous smoking measures so derived was categorized at approximate quartile cutpoints from the control distribution.

Statistical methods
In the analyses, we aimed to quantify the associations between key measures of smoking exposure (duration, intensity, dose, and time since quitting) and the risks of EAC, GEJAC, and ESCC. In particular, we sought to estimate the independent effect of each dimension of smoking after also considering the effects of the other smoking dimensions. Our approach was first to fit standard logistic models using single measures of smoking separately, adjusted for the potentially confounding effects of nonsmoking factors, similar to previous studies that investigated smoking and esophageal cancer (2, 3, 8–10). We then fitted more complex models in which multiple smoking measures were modeled simultaneously to assess which of the measures, if any, were most strongly associated with cancer risk (11, 15). Finally, we assessed the effect of smoking cessation on cancer risk by adding the term "time since quitting" to models containing other dimensions of smoking exposure.

Risk estimates for individual smoking measures.
For single smoking measures, we used multivariate logistic regression (GENMOD procedure) in SAS, version 9.1 (SAS Institute, Inc., Cary, North Carolina) to calculate odds ratios and 95 percent confidence intervals, using never smokers as the reference category. Statistical significance was determined using a two-sided test at {alpha} = 0.05. Trend tests for ordinal categorical variables were performed using median values for each category as continuous values in the model and were restricted to exposed groups only. We adjusted for the matching variables (sex, age) as well as educational level, average weekly alcohol consumption, body mass index, aspirin use, and frequency of heartburn or reflux 10 years prior to participation.

Risk estimates for qualitative and individual quantitative smoking measures.
In the next step, we sought to simultaneously measure the qualitative effect of smoking and the dose effect for each dimension. We fitted a series of models which included a smoking indicator term (never, ever) as well as a separate term for each specific dimension of smoking. To allow the model to simultaneously estimate risks of smoking overall, together with risks for each dimension, we transformed the continuous measures by subtracting their respective mean values from the original values among smokers and allocating a value of zero for never smokers (a transformation known as "centering") (11, 16). Thus, the transformed continuous variables had a value of zero for both never smokers and "average" smokers; for all other smokers, the value of the transformed variable was the deviation from the mean. The indicator variable was estimated at the same value (zero) for both never smokers and average smokers; the dose effect was estimated from smokers only and depended on the extent to which their risk varied linearly (on the logit scale) with their increment in smoking above or below the mean. Thus, when interpreting risk estimates from these models, it is important to recognize that the reference value is the mean for the dimension of smoking being assessed.

Risk estimates for qualitative and multiple simultaneous quantitative smoking measures.
In the final model, we added other continuous smoking variables to quantify the dose-response for each dimension of smoking while adjusting for the others. The model-building strategy was based on Akaike's information criterion; models with minimum Akaike values were considered the best fit for each outcome.

All regression analyses were performed on participants with complete data for smoking and other confounding factors so that Akaike's information criterion was comparable across the nested (or nonnested) models. Altogether, 121 participants (46 controls, 21 cases with EAC, 22 cases with GEJAC, and 32 cases with ESCC) were excluded from the analysis because of missing data. Among those, 89 had data on smoking (30 percent nonsmokers, 44 percent ex-smokers, and 26 percent current smokers).

In the risk factor models for each outcome, we checked for nonlinearity in the dose effect of continuous smoking variables by means of generalized additive logistic models (17), using R software (CRAN package mgcv) (18, 19). Cubic splines fixed at 3 degrees of freedom (df) were used to test for nonlinear effects; nonlinear terms were retained only if they significantly improved model fit (17).


   RESULTS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
DISCUSSION
 References
 
Risk estimates for individual smoking measures
Table 1 presents the risk estimates for EAC, GEJAC, and ESCC for each dimension of smoking fitted as a categorical variable, adjusted for other nonsmoking confounding variables. Risk estimates for smoking measures were modestly attenuated after adjustment for confounding variables (particularly acid reflux and body mass index for EAC and GEJAC and alcohol for ESCC) (data not shown). Compared with never smokers, current smokers had significantly increased risks of ESCC and GEJAC and a greater than twofold elevation in risk for EAC. Ex-smokers had more than a twofold increased risk of ESCC and significant 40–70 percent elevations for GEJAC and EAC. We found no association with age at starting smoking for any type of esophageal cancer; however, the range was limited. Longer durations of smoking were more strongly associated with cancer than shorter durations; significantly increased risks were observed for smoking durations greater than 25 years for EAC, for durations greater than 15 years for GEJAC, and for all durations for ESCC. There was no significant linear trend in the risk of EAC or GEJAC with increasing smoking intensity, whereas a significant trend was observed with increasing smoking intensity for ESCC. Total cumulative dose was also significantly associated with all subtypes, with generally higher risks being observed for ESCC than for EAC and GEJAC for a given category.


View this table:
[in this window]
[in a new window]

  		TABLE 1. Odds ratios for different histologic types of esophageal cancer according to various dimensions of tobacco smoking in a simple categorical analysis, Australia, 2002–2005

 
We assessed the effect of smoking cessation by comparing the risks of cancer between ex-smokers and never smokers. In this analysis, risks of EAC among ex-smokers remained significantly elevated until 20 years postcessation; for GEJAC and ESCC, risks among ex-smokers remained elevated for up to 30 years.

Effects of smoking duration, intensity, and cumulative dose
When we estimated the risks of cancer associated with incremental changes in each dimension of smoking modeled simultaneously with an indicator variable (ever/never smoker), the highest risks for each type of cancer were observed with the qualitative dimension "ever smoking" (table 2). The odds of ever smoking were elevated approximately 70 percent for EAC, 130 percent for GEJAC, and 180 percent for ESCC, regardless of which other dimensions of smoking were included in the model. After partitioning of the qualitative effect of ever smoking, risks for all types of esophageal cancer increased significantly with longer durations of smoking. We formally tested the effect of nonlinear terms for smoking duration but found no evidence that they were associated with EAC (p = 0.33), GEJAC (p = 0.32), or ESCC (p = 0.12). Increasing intensity of smoking as a linear term in the model was associated with increased risk of ESCC but not EAC or GEJAC. Compared with the effects observed for smoking duration, we found weaker associations and poorer-fitting models when cumulative dose was assessed for EAC and GEJAC; for ESCC, the cumulative dose term provided a fit similar to those for duration and intensity.


View this table:
[in this window]
[in a new window]

  		TABLE 2. Odds ratios for different histologic types of esophageal cancer according to various dimensions of tobacco smoking in simultaneous modeling of qualitative and individual quantitative smoking measures, Australia, 2002–2005

 
When linear terms for duration and intensity were included simultaneously in the model, duration remained significantly associated with EAC (odds ratio (OR) = 1.20, 95 percent confidence interval (CI): 1.06, 1.36) and GEJAC (OR = 1.17, 95 percent CI: 1.05, 1.31) (table 3), but intensity showed no improvement in the model goodness of fit in comparison with the model that included a term only for duration (data not shown). Although the linear effect of smoking intensity showed no significant association with GEJAC, the effect became significant (p = 0.02) when intensity was included as a nonlinear function in the model (figure 1, part A) and improved the model's goodness of fit significantly (p = 0.01). The risk of GEJAC increased steadily by 65–70 percent for each additional increment of 10 cigarettes/day, plateaued at 25 cigarettes/day, and then declined for persons who smoked more than 30 cigarettes/day.


View this table:
[in this window]
[in a new window]

  		TABLE 3. Odds ratios for different histologic types of esophageal cancer according to various dimensions of tobacco smoking in simultaneous modeling, Australia, 2002–2005

 

Figure 1
View larger version (12K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		FIGURE 1. Relations of intensity of smoking and its dose-pattern effect with risks of (A) gastroesophageal junction adenocarcinoma and (B) esophageal squamous cell carcinoma, Australia, 2002–2005. Results were obtained using generalized additive logistic models with a 3-df cubic spline function. The solid line represents the odds ratio, and the dashed line represents its 95% confidence interval. The reference value for the estimate was fixed around the mean of the continuous variable; hence, the odds ratio for an association between two intensity values (x-axis) can be calculated as the ratio of the two corresponding odds ratios (y-axis, log scale).

 
In contrast, when linear terms for duration and intensity were modeled simultaneously for ESCC, both retained sizeable and significant risk estimates and resulted in a better-fitting model. As for GEJAC, modeling smoking intensity as a nonlinear function significantly improved the model fit (p = 0.05); the odds of ESCC increased significantly by 50 percent for every extra 10 cigarettes/day, plateaued at 40 cigarettes/day, and declined slightly for persons who smoked more than 60 cigarettes/day.

Effects of smoking cessation
After partitioning of the qualitative effect of ever smoking, time since quitting was associated with significant risk reductions for all three subtypes on the order of 20 percent per 10 years of cessation, in comparison with current smokers. The risk estimate remained essentially unchanged regardless of which other dimensions of smoking were included in the models (table 3). After adjusting for intensity as a nonlinear function in the model, the effect of time since quitting was attenuated slightly for GEJAC (OR = 0.83, 95 percent CI: 0.74, 0.93) and ESCC (OR = 0.85, 95 percent CI: 0.75, 0.98), but it remained significant. The risk observed in relation to time since quitting declined in a linear fashion for all three sites (p = 0.86, p = 0.18, and p = 0.57 for EAC, GEJAC, and ESCC, respectively, when tested against linearity) (figure 2). Smoking intensity remained a significant risk factor for GEJAC and ESCC independently of time since quitting (table 3). Conversely, risk estimates for duration of smoking were attenuated to the null for all three cancers when the term for time since quitting was also included in the model.


Figure 2
View larger version (8K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		FIGURE 2. Relations of time since quitting smoking and its dose-pattern effect with risks of (A) esophageal adenocarcinoma, (B) gastroesophageal junction adenocarcinoma, and (C) esophageal squamous cell carcinoma, Australia, 2002–2005. Results were obtained using generalized additive logistic models with a 3-df cubic spline function. The solid line represents the odds ratio, and the dashed line represents its 95% confidence interval. The reference value for the estimate was fixed around the mean of the continuous variable; hence, the odds ratio for an association between two values for time since quitting (x-axis) can be calculated as the ratio of the two corresponding odds ratios (y-axis, log scale).

 

   DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
References
 
We have confirmed that smoking is a sizeable and significant risk factor for each histologic subtype of esophageal cancer but have shown qualitative and quantitative differences in the associations by subtype. Thus, while duration and intensity of smoking were both independently associated with GEJAC and ESCC, we found no evidence that smoking intensity predicted risk of EAC. Finally, these analyses demonstrated the benefits of smoking cessation in reducing the risks of all three types of cancer, after accounting for the other qualitative and quantitative effects of smoking on cancer risk.

The role of smoking (and cessation) in the development of EAC has been of particular interest because of the substantial increase in incidence among Western populations during a period when the overall prevalence of smoking has declined. Risk estimates derived from our preliminary categorical analyses accord with previous estimates (2, 3, 8), except that in all earlier studies, investigators reported significant dose-response trends for every dimension of smoking assessed. Unlike investigators in previous studies, we found no association with age of smoking onset for any of the cancers or with smoking intensity for EAC. One likely explanation for these important differences is that earlier studies included "never smokers" as the reference category when testing for trend, which tends to overestimate the dose effect (11). Furthermore, to the best of our knowledge, previous investigations of smoking and esophageal cancer have not assessed the independent effect of each smoking dimension by simultaneously modeling other smoking dimensions; hence, previous dimension-specific estimates may have been confounded by other aspects of smoking exposure.

One strategy for addressing the complexity of the multidimensional nature of smoking exposure is to fit models that include both indicator terms for ever smoking and transformed variables for the continuous dimensions of smoking (11). Using this approach, our data showed that smoking duration is the key determinant of risk for adenocarcinoma, whereas intensity and duration together determine the risk of squamous cell carcinoma. These findings strongly suggest that the mechanisms by which EACs and ESCCs are induced in smokers differ. An overall stronger effect of smoking for squamous cell carcinoma as compared with adenocarcinoma has also been observed in lung cancers (4, 20, 21). However, these studies have not directly compared the effects of duration and intensity among the different histologic subtypes.

By analyzing the effects of the various dimensions of smoking using generalized additive models, we were able to explore possible nonlinear effects of these exposures on esophageal cancer risk. For most measures, we found no evidence that models incorporating nonlinear terms explained the association with esophageal cancer risk any better than linear models. However, for GEJAC and ESCC, we found that including a nonlinear term for smoking intensity produced significantly better-fitting models than those including a linear term. Nonlinear associations with smoking intensity have also been observed in lung cancer studies (4, 22) and suggest that the reduced risk of cancer among persons exposed to tobacco smoke at higher intensities may reflect increased DNA repair capacity (or reduced susceptibility to cancer) as opposed to misclassification or other sources of error.

Why should histologic subtypes of esophageal cancer have differing associations with patterns of smoking? Few directly applicable data are available; however, analogies in lung cancer epidemiology may offer insights. Similarly to esophageal cancer, there has been an increase in the ratio of adenocarcinoma to squamous cell carcinoma for lung cancer (23, 24); this has been attributed to the introduction of filter-tipped and "low-tar" cigarettes, which has altered the patterns of respiratory epithelial exposure to tobacco carcinogens (25, 26). Whether this pertains to esophageal cancer is not known. At the cellular level, it appears that the chromosomal aberrations caused by tobacco smoke cluster differently in adenocarcinomas and squamous cell carcinoma of the lung (27). We are not aware of comparable investigations for esophageal cancers relating to tobacco exposure, but they could be highly informative.

Host factors also play some role in determining susceptibility to the carcinogenic effects of tobacco smoke. At least one report has identified combinations of polymorphisms in DNA repair genes that are differently associated with adenocarcinomas and squamous cell carcinoma of the lung (28), suggesting that constitutional genotype may influence not only overall cancer risk but also the type and site of smoking-related cancers. Doecke et al.'s (29) recent finding that polymorphisms in the MGMT gene (which codes for a protein that repairs alkylating mutations arising from nitrosamines) are associated with increased risks of EAC but not GEJAC accords with this notion and provides some clues to the potential etiologic pathways of esophageal cancers.

From public health and clinical perspectives, our analyses offer hope that people who permanently cease smoking will significantly reduce their risk of all types of esophageal cancer. We estimated that the magnitude of the risk reduction was approximately 15–19 percent for every 10 years of smoking cessation. In our data set, smoking duration and time since quitting were correlated measures with effect sizes that were almost equal in opposite directions. When these factors were adjusted for each other, only time since quitting remained statistically significant, suggesting that this effect dominated among ex-smokers.

Strengths of our study included the large sample size, the population-based sampling frame, and the systematic collection of detailed smoking data. We minimized the possibility of biased recall by concealing study hypotheses from participants and interviewers. Our inferences regarding the effects of smoking on esophageal cancer were strengthened by the consistent differences in associations by histologic type, a circumstance unlikely to be explained by chance or differential reporting.

A limitation of our study was the low participation rate among controls, increasing the likelihood that our control sample was not representative of the population from which the cases arose. To assess the magnitude of possible bias, we compared smoking prevalence in our control group with that reported for the 2004 Australian National Health Survey (30), a representative survey of the Australian adult population. While the distribution of overall smoking status in controls was similar to that of the overall population, ex-smokers were somewhat overrepresented among our controls. We estimated the effects of potentially biased participation by imputation analysis and found that increases in risk estimates derived from the imputed data remained over twofold and significant for both current smokers and ex-smokers. Some degree of selection bias may have also beset the cases, although its direction and magnitude are difficult to estimate. Survival bias, where cases with long survival are overrepresented in the study sample, might have been present if smoking status were associated with survival among esophageal cancer patients. We compared survival for each group of case participants according to smoking status using mortality data collected subsequently and found no significant differences within any of the case groups, suggesting that any pertinent survivor bias was unlikely to account for the different patterns of risk between the groups. We also repeated our analyses after excluding the subset of cases identified through the cancer registries (for whom longer survival might have contributed to recruitment), with little difference in estimates being observed.

In summary, we found qualitative and quantitative differences in the association between dimensions of smoking exposure and EAC, GEJAC, and ESCC. Duration of smoking was a strong determinant of all three types of cancer, but intensity was associated only with risks of GEJAC and ESCC, and there was a significant nonlinear dose effect of intensity on risk of these cancers. Time since quitting was an independent predictor for all outcomes and was the most dominant dimension of smoking. Our findings emphasize the benefits of quitting smoking in reducing the risk of esophageal cancer, irrespective of how long or how heavily a person has smoked.


   ACKNOWLEDGMENTS
 
This study was supported by the Cancer Council Queensland and the National Health and Medical Research Council of Australia (NHMRC) (program 199600). D. C. W. and P. M. W. were supported by Senior Research Fellowships from the NHMRC and Cancer Council Queensland, respectively. N. P. was supported by a doctoral scholarship from the NHMRC. S. S. was supported by a doctoral scholarship from the Ministry of Health and Medical Education of the Islamic Republic of Iran.

The authors acknowledge Dr. Adrian Barnett (University of Queensland, Brisbane, Australia) for his help with R programming and his comments on the manuscript. Dr. Jay Lubin (National Cancer Institute, Bethesda, Maryland) provided thoughtful feedback on this work.

The funding bodies played no role in the design or conduct of the study; in the collection, management, analysis, or interpretation of the data; or in the preparation, review, or approval of the manuscript.

Conflict of interest: none declared.


   References
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 References
 

  1. Vaughan TL, Davis S, Kristal A, et al. Obesity, alcohol, and tobacco as risk factors for cancers of the esophagus and gastric cardia: adenocarcinoma versus squamous cell carcinoma. Cancer Epidemiol Biomarkers Prev (1995) 4:85–92.[Abstract]
  2. Gammon MD, Schoenberg JB, Ahsan H, et al. Tobacco, alcohol, and socioeconomic status and adenocarcinomas of the esophagus and gastric cardia. J Natl Cancer Inst (1997) 89:1277–84.[Abstract/Free Full Text]
  3. Lagergren J, Bergstrom R, Lindgren A, et al. The role of tobacco, snuff and alcohol use in the aetiology of cancer of the oesophagus and gastric cardia. Int J Cancer (2000) 85:340–6.[CrossRef][Web of Science][Medline]
  4. Lubin JH, Caporaso NE. Cigarette smoking and lung cancer: modeling total exposure and intensity. Cancer Epidemiol Biomarkers Prev (2006) 15:517–23.[Abstract/Free Full Text]
  5. Hashibe M, Boffetta P, Janout V, et al. Esophageal cancer in Central and Eastern Europe: tobacco and alcohol. Int J Cancer (2007) 120:1518–22.[CrossRef][Web of Science][Medline]
  6. Lindblad M, Rodriguez LAG, Lagergren J. Body mass, tobacco and alcohol and risk of esophageal, gastric cardia, and gastric non-cardia adenocarcinoma among men and women in a nested case-control study. Cancer Causes Control (2005) 16:285–94.[CrossRef][Web of Science][Medline]
  7. Cheng KK, Sharp L, McKinney PA, et al. A case-control study of oesophageal adenocarcinoma in women: a preventable disease. Br J Cancer (2000) 83:127–32.[CrossRef][Web of Science][Medline]
  8. Wu AH, Wan P, Bernstein L. A multiethnic population-based study of smoking, alcohol and body size and risk of adenocarcinomas of the stomach and esophagus (United States). Cancer Causes Control (2001) 12:721–32.[CrossRef][Web of Science][Medline]
  9. Freedman ND, Abnet CC, Leitzmann MF, et al. A prospective study of tobacco, alcohol, and the risk of esophageal and gastric cancer subtypes. Am J Epidemiol (2007) 165:1424–33.[Abstract/Free Full Text]
  10. Anderson LA, Watson RG, Murphy SJ, et al. Risk factors for Barrett's oesophagus and oesophageal adenocarcinoma: results from the FINBAR study. World J Gastroenterol (2007) 13:1585–94.[Web of Science][Medline]
  11. Leffondré K, Abrahamowicz M, Siemiatycki J, et al. Modeling smoking history: a comparison of different approaches. Am J Epidemiol (2002) 156:813–23.[Abstract/Free Full Text]
  12. Leffondré K, Abrahamowicz M, Xiao Y, et al. Modelling smoking history using a comprehensive smoking index: application to lung cancer. Stat Med (2006) 25:4132–46.[CrossRef][Web of Science][Medline]
  13. Whiteman DC, Sadeghi S, Pandeya N, et al. Combined effects of obesity, acid reflux and smoking on the risk of adenocarcinomas of the oesophagus. Gut (2008) 57:173–80.[Abstract/Free Full Text]
  14. Olsen CM, Bain CJ, Jordan SJ, et al. Recreational physical activity and epithelial ovarian cancer: a case-control study, systematic review, and meta-analysis. Cancer Epidemiol Biomarkers Prev (2007) 16:2321–30.[Abstract/Free Full Text]
  15. Rachet B, Siemiatycki J, Abrahamowicz M, et al. A flexible modeling approach to estimating the component effects of smoking behavior on lung cancer. J Clin Epidemiol (2004) 57:1076–85.[CrossRef][Web of Science][Medline]
  16. Kraemer HC, Blasey CM. Centering in regression analyses: a strategy to prevent errors in statistical inference. Int J Methods Psychiatr Res (2004) 13:141–51.[CrossRef][Web of Science][Medline]
  17. Hastie TJ, Tibshirani RJ. Generalized additive models. (Monographs on statistics and applied probability, no. 43). (1990) London, United Kingdom: Chapman & Hall Ltd.
  18. RDevelopment Core Team. R: a language and environment for statistical computing (version 2.3.1). (ISBN 3-900051-07-0). Vienna, Austria: R Foundation for Statistical Computing, 2006. (http://www.R-project.org).
  19. Wood SN. Generalized additive models: an introduction with R. (Texts in statistical science). (2006) Boca Raton, FL: CRC Press.
  20. Khuder SA. Effect of cigarette smoking on major histological types of lung cancer: a meta-analysis. Lung Cancer (2001) 31:139–48.[CrossRef][Web of Science][Medline]
  21. Jedrychowski W, Becher H, Wahrendorf J, et al. Effect of tobacco smoking on various histological types of lung cancer. J Cancer Res Clin Oncol (1992) 118:276–82.[CrossRef][Web of Science][Medline]
  22. Lubin JH, Caporaso N, Wichmann HE, et al. Cigarette smoking and lung cancer: modeling effect modification of total exposure and intensity. Epidemiology (2007) 18:639–48.[CrossRef][Web of Science][Medline]
  23. Li X, Mutanen P, Hemminki K. Gender-specific incidence trends in lung cancer by histological type in Sweden, 1958 –1996. Eur J Cancer Prev (2001) 10:227–35.[CrossRef][Web of Science][Medline]
  24. Blizzard L, Dwyer T. Lung cancer incidence in Australia: impact of filter-tip cigarettes with unchanged tar yields. Int J Cancer (2002) 97:679–84.[CrossRef][Web of Science][Medline]
  25. Djordjevic MV, Hoffmann D, Hoffmann I. Nicotine regulates smoking patterns. Prev Med (1997) 26:435–40.[CrossRef][Web of Science][Medline]
  26. De Stefani E, Boffetta P, Ronco AL, et al. Squamous and small cell carcinomas of the lung: similarities and differences concerning the role of tobacco smoking. Lung Cancer (2005) 47:1–8.[CrossRef][Web of Science][Medline]
  27. Pan H, Califano J, Ponte JF, et al. Loss of heterozygosity patterns provide fingerprints for genetic heterogeneity in multistep cancer progression of tobacco smoke-induced non-small cell lung cancer. Cancer Res (2005) 65:1664–9.[Abstract/Free Full Text]
  28. Popanda O, Schattenberg T, Phong CT, et al. Specific combinations of DNA repair gene variants and increased risk for non-small cell lung cancer. Carcinogenesis (2004) 25:2433–41.[Abstract/Free Full Text]
  29. Doecke J, Zhao ZZ, Pandeya N, et al. Polymorphisms in MGMT and DNA repair genes and the risk of esophageal adenocarcinoma. Int J Cancer 2008; Apr 3 [Epub ahead of print].
  30. Australian Bureau of Statistics. National Health Survey: users' guide—electronic publication, 2004 –05. (Catalogue no. 4363.0.55.001). Canberra, Australia: Australian Bureau of Statistics, 2006. (http://www.abs.gov.au/ausstats/abs@.nsf/mf/4363.0.55.001).

Add to CiteULike CiteULike   Add to Connotea Connotea   Add to Del.icio.us Del.icio.us    What's this?


This article has been cited by other articles:


Home page
 GutHome page
J Steevens, L J Schouten, R A Goldbohm, and P A van den Brandt
Alcohol consumption, cigarette smoking and risk of subtypes of oesophageal and gastric cancer: a prospective cohort study
Gut, January 1, 2010; 59(01): 39 - 48.
[Abstract] [Full Text] [PDF]

Home page
 GutHome page
N Pandeya, P M Webb, S Sadeghi, A C Green, D C Whiteman, and for the Australian Cancer Study
Gastro-oesophageal reflux symptoms and the risks of oesophageal cancer: are the effects modified by smoking, NSAIDs or acid suppressants?
Gut, January 1, 2010; 59(01): 31 - 38.
[Abstract] [Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow All Versions of this Article:
168/1/105    most recent
kwn091v1
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrow Search for citing articles in:
ISI Web of Science (8)
Right arrowRequest Permissions
Right arrow Disclaimer
Google Scholar
Right arrow Articles by Pandeya, N.
Right arrow Articles by Whiteman, D. C.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Pandeya, N.
Right arrow Articles by Whiteman, D. C.
Social Bookmarking
Add to CiteULike   Add to Connotea   Add to Del.icio.us  
What's this?