|Home | About | Journals | Submit | Contact Us | Français|
To evaluate the dose–response relationship between cigarette smoking and pancreatic cancer and to examine the effects of temporal variables.
We analyzed data from 12 case–control studies within the International Pancreatic Cancer Case–Control Consortium (PanC4), including 6507 pancreatic cases and 12 890 controls. We estimated summary odds ratios (ORs) by pooling study-specific ORs using random-effects models.
Compared with never smokers, the OR was 1.2 (95% confidence interval [CI] 1.0–1.3) for former smokers and 2.2 (95% CI 1.7–2.8) for current cigarette smokers, with a significant increasing trend in risk with increasing number of cigarettes among current smokers (OR = 3.4 for ≥35 cigarettes per day, P for trend <0.0001). Risk increased in relation to duration of cigarette smoking up to 40 years of smoking (OR = 2.4). No trend in risk was observed for age at starting cigarette smoking, whereas risk decreased with increasing time since cigarette cessation, the OR being 0.98 after 20 years.
This uniquely large pooled analysis confirms that current cigarette smoking is associated with a twofold increased risk of pancreatic cancer and that the risk increases with the number of cigarettes smoked and duration of smoking. Risk of pancreatic cancer reaches the level of never smokers ~20 years after quitting.
Cigarette smoking is the best established risk factor for pancreatic cancer [1, 2]. A meta-analysis of 82 cohort and case–control studies published between 1950 and 2007  reported a summary relative risk (RR) of pancreatic cancer of 1.7 (95% confidence interval [CI] 1.6–1.9) for current smokers and of 1.2 (95% CI 1.1–1.3) for former smokers. It also showed that the risk persisted for up to 10 years after quitting smoking, although no detailed analysis of the dose– and duration–risk relations was conducted. In the International Pancreatic Cancer Cohort Consortium nested case–control study , that included 1481 cases and 1539 controls, the RR was 1.1 (95% CI 0.9–1.3) for former smokers and 1.8 (95% CI 1.4–2.3) for current smokers. Significant trends in risk were observed with increased number of cigarettes smoked and duration of exposure, the RR being 1.75 for 30 or more cigarettes smoked per day and 2.1 for 50 or more years of smoking, whereas the RR for those who had quit smoking for >15 years was similar to that of never smokers.
To further evaluate the dose–response relationship between cigarette smoking and pancreatic cancer and the role of various temporal factors, such as age at starting and time since stopping, we analyzed the original data from a series of case–control studies within the International Pancreatic Cancer Case–Control Consortium (PanC4) [5, 6]. This uniquely large dataset allowed us to investigate in detail cigarette smoking on pancreatic cancer, with careful adjustment for major potential confounding factors for pancreatic cancer.
The PanC4 identified 12 case–control studies (including the unpublished Louisiana State University study) of pancreatic cancer that collected data on cigarette smoking using structured questionnaires [7–18]. Eight studies [7–14] (including the unpublished Louisiana State University study) were conducted in North America, two in Europe [15, 16], and one in China , and one was the International Agency for Research on Cancer-coordinated Surveillance of Environmental Aspects Related to Cancer in Humans (SEARCH) study from Canada, Europe, and Australia . A summary description of the individual studies is presented in Table Table11.
The present pooled analysis included a total of 6507 cases of adenocarcinoma of the exocrine pancreas and 12 890 controls. The data included in the pooled analysis may differ slightly from those in published reports of the individual studies due to missing data for relevant variables. In all studies, cases and controls were interviewed in person, with the exception of the Ontario, Canada study that used self-administered questionnaires and included 63 case proxy respondents ; the SEARCH study , where proxy interviews were conducted for 474 cases and 332 controls; and the Shanghai study , where 155 cases and 150 controls were proxy interviewed.
The original datasets were restructured either by the original study investigators or by the central coordinators using a uniform format for data harmonization. In addition to the smoking-related data, for each study, we considered individual data on sociodemographic characteristics, anthropometric measures, alcohol consumption, and history of diabetes and pancreatitis.
All studies in this pooled analysis provided information about cigarette smoking status (never, former, and current smoker), number of cigarettes smoked per day, duration of smoking, age at start smoking, and years since quitting or age at quitting smoking for former smokers. Though questions about cigarette smoking were similar across studies, we conducted a careful and detailed examination of the comparability of smoking-related questions to harmonize the data from the multiple studies included in this pooled analysis.
For the present analyses, ever cigarette smokers were defined as participants who had smoked at least 100 cigarettes in their lifetime [7–9, 11, 13, 14, 17] or more than one cigarette per day for at least 1 year [15, 16, 18]. In the National Cancer Institute (NCI), information was only available for regular smokers, i.e. those who smoked at least one cigarette per day for at least 6 months . Former cigarette smokers were defined as those who had quit smoking for at least 1 year before interview in all studies.
To estimate the association between cigarette smoking and pancreatic cancer risk, we used a two-stage modeling approach . In the first stage, we assessed the association between cigarette smoking and pancreatic cancer for each study by estimating the odds ratios (ORs) and the corresponding 95% CIs using multivariable unconditional logistic regression models . These models included terms for age (<50, 50–54, 55–59, 60–64, 65–69, 70–74, ≥75 years), sex, education (≤8th grade, 9th–11th grade, 12th grade or high school graduates, some college or college graduates, ≥1 year of graduate school), race/ethnicity (non-Hispanic white, Hispanic, non-Hispanic black, other), body mass index (BMI; <20, 20 to <25, 25 to <30, ≥30 kg/m2), history of diabetes (≥1 year before diagnosis/interview), alcohol consumption (never drinkers, 1–6 drinks per day, ≥6 drinks per day), and study center for multicentric studies. In the second stage of the analysis, summary estimates were computed using random-effects models , weighting study-specific ORs by the inverse of the sum of their variance and the estimated between-study variance component. Heterogeneity between studies was evaluated using the Q test statistic . To test for the significance of linear trends in pancreatic cancer risk across levels of cigarette smoking, we first estimated the trends in each study and then used the Wald test to estimate the P value of the summary variable from the random-effects models .
To investigate whether the effect of cigarette smoking was homogeneous within strata of selected covariates, we conducted analyses stratified by sex, age (<65, ≥65 years), alcohol consumption (never drinkers or 1–4 drinks per day, ≥4 drinks per day), race/ethnicity (non-Hispanic White, other races), study area (North America, Europe, other race/ethnicity), source of controls (population based, hospital based), type of respondents (in person, proxy). Significance of the heterogeneity across individual strata was assessed using a χ2 statistic .
We also conducted a sensitivity analysis to evaluate the undue influence of a study on the overall summary estimates by excluding one study at a time from the pooled analysis. Moreover, we carried out a cumulative meta-analysis to determine whether the association between cigarette smoking and pancreatic cancer risk changed over time.
In addition to the two-stage analysis, we conducted an aggregate analysis where data from all studies were pooled into a single large dataset . The association between cigarette smoking and pancreatic cancer risk was determined using multivariable unconditional multiple logistic regression models . Models included terms for study area as well as for the confounding factors from the study-specific models and the study–confounder interactions. The results were not substantially different from those obtained in the two-stage analysis approach and therefore are not reported here.
Table Table22 shows the distribution of 6507 pancreatic cancer cases and 12 890 controls by sex, age, and other potential confounding factors. Cases and controls have a similar sex distribution; cases were somewhat older than controls, were more frequently non-Hispanic White, had a higher level of education, a higher BMI, and more frequently reported a history of diabetes and pancreatitis.
The pooled ORs for pancreatic cancer according to cigarette smoking habits are given in Table Table3.3. Compared with never smokers, the OR was 1.40 (95% CI 1.24–1.55) for ever cigarette smokers, 1.17 (95% CI 1.02–1.34) for former cigarette smokers, and 2.20 (95% CI 1.71–2.83) for current cigarette smokers. A significant trend in risk was observed with increased number of cigarettes smoked (OR = 3.4, 95% CI 2.4–4.9 for ≥35 cigarettes per day, P for trend <0.0001). Among current smokers, the risk increased with increased duration of cigarette smoking for up to 40 years of smoking (OR = 2.43, 95% CI 1.91–3.09) but did not increase further after 40 years. No trend in risk was observed for age at starting cigarette smoking in current smokers, whereas a significant decreasing trend in risk was found with increased time since quitting cigarette smoking. After ~20 years, risk estimates were not different from nonsmokers (OR = 0.98). Sensitivity analyses showed that no single study unduly influenced the magnitude or the statistical significance of these summary estimates.
The cumulative meta-analysis for pancreatic cancer risk in current cigarette smokers showed a trend of increasing risk according to the year of publication: the OR for current versus never smokers was 1.74 in the initial study published in 1994, 1.75 in the studies published up to 2007, and rose to 2.2 when studies published between 2007 and 2010 were added (Figure (Figure11).
A forest plot of the study-specific and the pooled ORs for pancreatic cancer risk for ever versus never cigarette smokers is presented in Figure Figure2.2. The corresponding forest plots for numbers of cigarettes smoked per day among current smokers are given in Figure Figure3.3. The pooled estimate for ever cigarette smokers as compared with never smokers was 1.40, with significant heterogeneity in ORs across studies (P = 0.003). Similarly, for current smokers of <15 (Figure (Figure3A),3A), 15–24 (Figure (Figure3B),3B), and ≥25 (Figure (Figure3C)3C) cigarettes per day, pooled estimates were significantly elevated, although between-study heterogeneity was observed for each level of cigarette smoking.
The association between number of cigarettes smoked and pancreatic cancer risk was further assessed in analyses stratified by sex, age, alcohol drinking, race/ethnicity, study area, source of controls, and type of respondents (Table (Table4).4). The association appeared somewhat stronger—though not significantly—in women, in participants <65 years, and in proxy respondents; no meaningful differences in risk estimates were observed across strata of other covariates considered.
This uniquely large collaborative pooled analysis within the PanC4 allowed us to provide more accurate estimates of the relationship between cigarette smoking and pancreatic cancer risk. Results from our analyses confirm that current cigarette smoking is associated with a twofold increased risk of pancreatic cancer and that the risk increases with increasing number of cigarettes smoked and duration of smoking. A 20% excess risk of pancreatic cancer was found among former smokers, which declines with time since quitting, and reached the level of never cigarette smokers ~20 years after quitting.
The increased pancreatic cancer risk in current cigarette smokers in our data is consistent with that of a previous meta-analysis and a pooled analysis [3, 4] and with the results from a subsequent study of the European Prospective Investigation into Cancer and Nutrition cohort  that reported a 70% increased risk (95% CI 1.4–2.2) for current smokers on the basis of 524 pancreatic cancer cases among 465 910 participants. The OR estimate for current cigarette smokers from our data was slightly higher than that reported in previous investigations [3, 4, 23]. This can be explained by the better distinction between current and former smokers that was possible in our analysis, but not in the studies included in the meta-analysis by Iodice et al. , as well as in prospective studies [4, 23], where smoking habits are generally assessed at the time of recruitment or last interview, and may have changed in subsequent years , i.e. some current smokers at baseline may have quit before diagnosis.
The significant dose–risk association with increasing number of cigarettes smoked is consistent with that reported in the pooled analysis of cohort studies . However, case–control studies data allowed us to assess this association specifically among current smokers, something that was not possible in cohort studies where the effect pertained to ever smokers.
With reference to duration of smoking, we observed that the risk of pancreatic cancer increased in relation to the years of smoking, up to 40 years of smoking. This confirms the importance of long-term duration of smoking as a risk factor for pancreatic cancer [1, 4].
The results of our study also confirmed the decline in risk of pancreatic cancer with increasing time since quitting cigarette smoking [3, 4]. More specifically, our large population and detailed data on smoking allowed us to confirm that after 20 years of smoking cessation, risk of pancreatic cancer approaches that of never smokers. This result was in close agreement with the findings from the International Pancreatic Cancer Cohort Consortium .
The PanC4 study had a number of strengths. It included original and detailed data about cigarette smoking for >6000 pancreatic cancer cases and >12 000 controls, which provided a unique opportunity to investigate and quantify accurately the dose- and duration–risk relationships, and among former smokers, the pattern of risk with years since quitting. Our study included a relatively large number of heavy and long-term smokers as well as a large number of former smokers, increasing our ability to examine smoking behaviors in greater detail than previous studies. We were able to uniformly and carefully account for study design variables and potential confounding factors for pancreatic cancer, including education, BMI, history of diabetes and pancreatitis, and heavy alcohol consumption. We also conducted stratified analyses by selected covariates and showed that our risk estimates were consistent across strata of sex, age, race, study area, and alcohol consumption.
Although there was significant heterogeneity between the 12 studies included in our pooled analysis, this was not explained by sex, age, study areas, source of controls, or other selected covariates considered and was largely attributable to the Mayo  and Milan studies only . This may be due to different background risk levels in various populations, bias, or simply the play of chance. Because N-nitrosamines are considered the major tobacco carcinogens for the pancreas [25, 26], some of the heterogeneity also might be related to different N-nitrosamines yield of cigarettes from various countries [27, 28]. In the absence of comprehensive data, however, any inference on this issue remains speculative.
Both hospital-based and population-based case–controls are prone to potential bias. Hospital controls may have been admitted to hospital for conditions related to tobacco use that could lead to an underestimation of the true association, whereas population controls may have a lower participation of smokers that could result in an overestimation of risk. Although the results from one meta-analysis  showed that RR estimates were higher in hospital-based than in population-based case–control studies, our stratified analyses by source of controls did not support this, showing no consistent difference in risk estimates when using hospital or population controls. Tobacco consumption is frequently underreported , and this may have biased our risk estimates, particularly if misclassification of smoking differed between cases and controls. However, information on smoking habits in case–control studies has proven to be satisfactorily reliable . Furthermore, the similarities of our findings with those from cohort studies  argue against a major role of recall bias and misclassification. Although proxy respondents may have reported tobacco consumption less completely than participants, we found no evidence of stronger associations in studies based on in-person interviews versus those using proxy respondents.
There is no early diagnosis or effective chemotherapy for pancreatic cancer . Most patients cannot undergo curative surgery, thus, even in the optimal series, 5-year survival is <5% . Consequently, primary prevention is the only way to reduce pancreatic cancer, and control of tobacco smoking is the key measure since it could avoid 15%–25% of pancreatic cancers in various populations [4, 16, 33].
The Louisiana State University study was supported by the Louisiana Board of Regents Millennium Trust Health Excellence Fund [project 5: HEF (2000–2005, Genetics Studies in the Acadian Population)]. The Pancreatic Cancer Family Registry at Memorial Sloan–Kettering Cancer Center has been supported by the Prevention, Control, and Population Research Goldstein Award, the Society of Memorial Sloan–Kettering Cancer Center, and the Geoffrey Beene Cancer Research Fund. The National Cancer Institute study was supported by the Intramural Research Program of the National Institute of Health, National Cancer Institute, Division of Cancer Epidemiology and Genetics (contract numbers: N01-CP-51090, N01-CP-51089, N01-CP-51092, N01-CP-05225, N01-CP-31022, N01-CP-05227). The University of California, San Francisco (UCSF) study work was supported in part by National Cancer Institute grants (CA59706, CA108370, CA109767, CA89726, CA098889 to EAH, PI) and by the Rombauer Pancreatic Cancer Research Fund. Cancer incidence data collection in the UCSF study was supported by the California Department of Public Health, the National Cancer Institute's Surveillance, Epidemiology and End Results program contract N01-PC-35136 awarded to the Northern California Cancer Center. The Yale Connecticut Study was supported by National Cancer Institute grant (5R01-CA098870 to HAR, PI). The Ontario Pancreas Cancer Study was supported by grants from the National Institutes of Health (R01 CA97075, as part of the PACGENE consortium), the Lustgarten Foundation for Pancreatic Cancer Research, and the Ontario Cancer Research Network. The Italian and Milan studies were supported by the Italian Association for Cancer Research (AIRC). The Montreal investigation in the Surveillance of Environmental Aspects Related to Cancer in Humans study was supported by the Cancer Research Society, the Toronto contribution was supported by the National Cancer Institute of Canada, and The Netherlands contribution was supported by the Dutch Ministry of Public Health, Welfare and Sports (formerly Welfare, Health and Culture).
The authors declare no conflicts of interest.
The authors thank Mrs Ivana Garimoldi for editorial assistance.