Alcohol consumption, cigarette smoking and the risk of subtypes of head-neck cancer: results from the Netherlands Cohort Study

Design and study population

The present study was conducted within the NLCS, which started in September 1986 with the inclusion of 120,852 participants, aged 55-69 years from 204 Dutch municipal population registries [19].

For data processing and analysis, the case-cohort design was used for reasons of efficiency [20]. Cases were derived from the total cohort, whereas the number of person-years at risk for the total cohort was estimated from a subcohort of 5000 persons, randomly sampled from the entire cohort at baseline.

Follow-up for cancer incidence was done by annual record linkage to the Netherlands Cancer Registry and the nationwide network and pathology registry [21]. The completeness of cancer follow-up is estimated to be ≥96% [22], and follow-up for vital status of the subcohort was nearly 100% complete after 17.3 years.

We excluded cohort members who reported to have prevalent cancer other than skin cancer at baseline, and cases and subcohort members with missing data on exposure or confounding variables. Only microscopically confirmed, first occurrences of squamous cell carcinomas – which include nearly all malignancies of the mouth, pharynx, and larynx [1, 3] – of the head and neck were included.

In total, 395 incident HNC cases and 4288 subcohort members were available for analysis (Figure 1). Of these cases, 110 were oral cavity cancer (ICD-O-3 C003-009, C020-C023, C030-C031, C039-C041, C048-C050, C060-C062, C068-C069), 83 oro-/hypopharyngeal cancer (C019, C024, C051-C052, C090-C091, C098-C104, C108-C109, C129-C132, C138-C139); 3 oral cavity, pharynx unspecified, or overlapping (C028-C029, C058-C059, C140-C142, C148), and 199 laryngeal cancer (C320-C329) cases, classified as proposed by Hashibe et al. [23], according to the International Classification of Diseases for Oncology (ICD-O-3) [24].

The NLCS has been approved by the Medical Ethics Committee of Maastricht University (Maastricht, The Netherlands).

Exposure information

At baseline, all cohort members completed a self-administered questionnaire, which included a 150-item food frequency questionnaire (FFQ) with detailed questions on alcohol consumption, smoking habits, and other cancer risk factors.

We asked about the habitual intake of alcohol during the year preceding the start of the study, measured by six items: (1) beer; (2) red wine; (3) white wine; (4) sherry and other fortified wines; (5) liquor types containing on average 16% alcohol; and (6) (Dutch) gin, brandy, and whisky. In addition, questions were asked about the frequency of consumption and the number of glasses consumed on each drinking occasion. For analysis, we combined (2), (3), and (4) into “wine”, and (5) and (6) into “liquor”. Total mean daily ethanol intake was calculated using the Dutch food-composition table [25]. On the basis of pilot study data, standard glass sizes were defined as 200 mL for beer, 105 mL for wine, and 45 mL for liquor, corresponding to 8g, 10g, and 13g of ethanol, respectively [26]. We also asked questions about the consumption of “beer” and “other alcoholic beverages” 5 years before baseline and selected participants with stable alcohol consumption to perform a sensitivity analysis [27]. Participants who indicated that they used alcoholic beverages never or less than once a month were considered abstainers.

We asked detailed information regarding cigarette smoking. Among others, questions were asked about whether the subject was a smoker at baseline; age at which they started and stopped smoking; the number of cigarettes smoked daily and the number of smoking years (excluding stopping periods). Based on these questions, the following variables were constructed for analysis: smoking status (never/former/current); current smoking (yes/no); frequency (cigarettes/day); duration (years); the number of pack-years; and time since smoking cessation (years). We also asked about cigar and pipe smoking and the use of smokeless tobacco. Participants who indicated they had never smoked cigarettes were considered never smokers.

The FFQ was validated against a 9-day diet record, and the Spearman correlation coefficient between the alcohol intake assessed by the questionnaire and that estimated by the diet record was 0.89 for all subjects and 0.85 for users of alcoholic beverages [28]. The reproducibility of the FFQ was assessed through annually repeated measurements in a subgroup of the subcohort and the test-retest correlation was 0.90 for alcohol intake; this correlation declined only 0.01-0.02 per year [29].

Data were key-entered and processed in a standardized manner, blinded with regard to case/subcohort status in order to minimize observer bias in coding and data interpretation.

Data analysis

Person-years at risk were calculated from baseline until diagnosis of HNC, death, emigration, loss to follow-up or end of follow-up (i.e. 31 December 2003), whichever occurred first.

Age (years) and sex were considered predefined confounders. The potential confounders considered were [3, 30, 31]: level of education, non-occupational physical activity, energy intake, coffee and tea consumption, intake of fruit, vegetables, fish, fat, red meat, meat products, and family history of head-neck cancer. Alcohol consumption and cigarette smoking were mutually adjusted in statistical models. A variable was considered a confounder if including it in the model changed the rate ratio (RR) for any of the cancer (sub-) types by >10%; according to this, none of the potential confounders was included in the final model.

The Cox proportional hazards model was used to estimate incidence RRs and corresponding 95% confidence intervals (CI) for alcohol consumption and cigarette smoking in multivariable adjusted case-cohort analyses. Analyses were done using the Stata 11.2 statistical software package (StataCorp, College Station, Texas, USA). Standard errors were calculated using the robust Huber-White sandwich estimator to account for additional variance introduced by sampling from the cohort; this method is equivalent to the variance-covariance estimator by Barlow [32]. The proportional hazards (PH) assumption was assessed using the scaled Schoenfeld residuals [33]. If there was an indication for violation of the assumption for a variable, we further investigated this by adding a time-varying covariate for that variable to the model.

We also analyzed beer, wine, and liquor consumption, adjusted for ethanol intake, to examine whether substances in alcoholic beverages, other than ethanol, have an effect on HNC-risk. In smoking analyses, different aspects of cigarette smoking were investigated and mutually adjusted for, in order to obtain a complete exposure model. The total number of cases that exclusively smoked cigar and/or pipe or used smokeless tobacco was too low (N < 10) to further analyze associations with HNC-risk.

When adjusting for smoking frequency, duration, or pack-years, we centered these continuous variables as proposed by Leffondré et al. [34]

Tests for linear dose-response trends were assessed by fitting ordinal exposure variables as continuous terms. To evaluate possible multiplicative interaction between categories of alcohol consumption and cigarette smoking, we estimated RRs of HNC overall and all HNC-subtypes for combinations of these exposures. The interaction was investigated by including cross-product terms in the model and performing a Wald test. Two-sided P values are reported throughout the article.

Tests for heterogeneity among HNC-subtypes were performed to investigate differences between HNC subtypes by a bootstrapping method developed for the case-cohort design [35]. For each bootstrap sample, X subcohort members were randomly drawn from the subcohort of X subjects and Y cases from the total of Y cases outside the subcohort, both with replacement, out of the dataset of X + Y observations. The logHRs were obtained from this sample using Stata’s competing risks procedure and recalculated for each bootstrap-replication. The confidence interval and P value of the differences in hazard ratio of the subtypes were then calculated from the replicated statistics. Each bootstrap analysis was based on at least 1,000 replications [36].

Alcohol consumption

After adjustment for total alcohol intake, consumption of beer, wine, and liquor was generally not significantly associated with HNC-risk. Beer consumption was, however, statistically significantly, positively associated with OHPC-risk (P trend = 0.03); liquor consumption was significantly associated with an increased risk of OCC (P trend = 0.03). Wine consumption was largely inversely associated – although not statistically significantly – with risk of HNC overall and HNC-subtypes.

Although risk rates clearly varied among HNC-subtypes, tests for heterogeneity did not show any significant risk differences, possibly due to low power.

Cigarette smoking

Regarding different aspects of cigarette smoking, after mutual adjustment, cigarette smoking status, frequency, and duration all remained statistically significantly associated with risk of HNC overall, OHPC, and LC (see Additional file 1). After additional adjustment for alcohol consumption (Table 3), most RRs between cigarette smoking status, frequency, duration and risk of HNC(-subtypes) slightly attenuated, but remained statistically significantly associated with increased risks.

Results regarding smoking cessation show that the risk of HNC overall and all subtypes diminished for smokers who stopped smoking since <10, 10 to <20, or ≥20 years, compared with current smokers (all P trend < 0.01) (Table 3). Nevertheless, compared with never smokers, RRs 20 years after smoking cessation were still elevated for HNC overall, OHPC, and LC, although not statistically significantly.

Despite considerable differences in risk rates among HNC-subtypes, tests for heterogeneity only showed statistically significant risk rates for duration of cigarette smoking (P < 0.001) and time since smoking cessation (P < 0.001).

Interaction between alcohol consumption and cigarette smoking

In HNC-subtypes, RRs were mostly increased as well when comparing participants smoking ≥ 20 cigarettes and drinking >15 g alcohol per day with never smokers consuming 0 to 15g alcohol per day, with the highest RR for OHPC (RR = 16.12, 95% CI 4.31-60.27), but no significant interaction was found, possibly due to low numbers in strata.

Alcohol consumption

Our results are in agreement with those of previous studies, showing alcohol consumption to be an independent risk factor for the development of HNC, with a strong, dose-response relationship [4, 9, 11–14, 17, 23, 37, 38]. Alcoholic beverages and acetaldehyde, the main metabolite of ethanol, are classified as a class I carcinogen [18]. It is plausible that alcohol – after being metabolized – acts both directly and indirectly in HNC carcinogenesis, the latter for example by acting as a solvent for other possible carcinogens, such as tobacco carcinogens [3, 39].

The differential risk among HNC-subtypes is consistent with other studies, in which LC was also least associated with alcohol consumption [8, 40, 41]. However, several other studies found OHPC being most associated with alcohol consumption, although sometimes in specific subgroups, as opposed to OCC in our study [11, 23, 41]. Nevertheless, the differential risk among HNC-subtypes is likely to be explained by the larynx having the least direct exposure to alcohol compared with the oral cavity and pharynx [39, 42]. The slightly increased RRs for alcohol consumption and LC may be due to inhalation of alcohol containing aerosols, silent aspiration, systemic effects, and possibly residual confounding.

After adjustment for total alcohol intake, we generally found similar risks between intake of beer, wine, liquor and HNC. These findings imply that ethanol itself probably is the most important factor in determining HNC-risk, rather than other substances in alcoholic beverages, which is in line with the results from other studies [3, 11, 42]. Consumption of wine was, however, generally inversely associated with HNC-risk, as was also shown in a pooled analysis [42], which could be due to residual confounding by a general healthier lifestyle of wine-consumers in our study population [3, 42, 43].

The significantly higher RRs between alcohol consumption and HNC risk in women as compared with men were seen earlier and could possibly be explained by women having stronger carcinogenic effects of alcohol at the same exposure level, suggesting possible gender-specific risk or protective factors [11].

Cigarette smoking

This study confirms the strong associations of cigarette smoking with increased risk of HNC overall and all subtypes [3, 5, 7, 10, 14, 23, 37, 41]. Among subtypes, however, OCC was least associated with cigarette smoking, and strongest associations were found with OHPC and LC. In addition, smoking status, frequency, and duration all appear to be of importance in the association between cigarette smoking and risk of HNC overall, OHPC, and LC. These results are generally consistent with previous reviews showing that cigarette smoking has a stronger effect on the larynx and/or pharynx than on the oral cavity [7, 8, 10, 23, 41]; in two meta-analyses, the larynx seemed to be clearly most susceptible to the effects of cigarette smoking [23, 41]. A possible explanation for this could be the aerodynamics of respiratory flow in the upper airway: this flow changes from laminar in the oral cavity to turbulent in the larynx, which may result in the larynx and pharynx having a higher exposure to inhaled air - and thus to cigarette smoke - than the oral cavity.

Finally, our study shows smoking cessation leads to decreased HNC-risks, which is in line with results from a recent pooled analysis as well [44].

Interaction between alcohol consumption and cigarette smoking

Our study confirms a multiplicative interaction between categories of alcohol consumption and cigarette smoking in HNC overall [9, 14, 16–18, 37, 38, 41]. The interaction effect between alcohol consumption and cigarette smoking is biologically plausible, since alcohol can act as a solvent for carcinogens in cigarette smoke and make the mucosa more permeable for these carcinogens; as a result, the carcinogenic properties of both factors are likely to be enhanced in the presence of one another [3, 39]. Still, in HNC-subtypes, we had low numbers of cases in strata, which probably resulted in limited power to detect a significant deviation from the multiplicative model.

Strengths and limitations

Important strengths of our study are the prospective character and completeness of follow-up. Our study is the second largest prospective cohort study investigating alcohol consumption and cigarette smoking on the risk of HNC overall and subtypes so far [9–15]. Furthermore, we were able to take into account data on smoking duration, and to investigate as well as adjust for several aspects of smoking behavior.

A possible limitation of our study is the single measurement of exposure data. Alcohol consumption and cigarette smoking were however extensively addressed in the questionnaire, with questions about lifetime exposure history of smoking and alcohol intake 5 years before baseline. It is however possible that participants who smoked at baseline in 1986 stopped smoking at some point during follow-up or changed their alcohol intake, and this may have led to bias due to misclassification. Furthermore, although our study includes a large number of cases, a lack of power is a possible explanation for finding non-significant results for some associations and the tests for heterogeneity.

We lack information on human papillomavirus (HPV) infection. HPV-infection is associated with HNC-risk [45, 46], but mainly with OHPC, in particular tonsil cancer and cancer of the base of the tongue. According to rates in our university medical center, only 25% of the diagnosed and treated oropharyngeal cancers between 1997 and 2003 were HPV-positive (all oropharyngeal cancer cases have been analyzed by p16-immunostaining and HPV16-specific fluorescence in situ hybridization (FISH), and – if FISH was negative – HPV-specific polymerase chain reaction). Moreover, the role of HPV in HNC-carcinogenesis is mainly of importance in young HNC-patients, and has increased since 1990 [47–49]. Since our participants were aged 55-69 years at baseline in 1986, we assume that the number of HPV-associated HNC cases in our cohort is low, and we expect potential bias due to possible misclassification to be very limited.

Other factors we did not take into account in our analyses are the use of drugs and oral hygiene. Although we investigated several potential confounders, residual confounding is still possible, but we presume this to be limited as well.

It might also be interesting to examine the RRs of HNC for smokers among non-drinkers and for drinkers among non-smokers. However, as the case numbers for these subgroups would be too small to analyze, we decided not to investigate this.

Finally, though we wanted to examine the role of alcohol consumption and cigarette smoking in HNC-subtypes, we did not investigate HNC located in the major salivary glands, nasal cavity, paranasal sinuses, and nasopharynx, because of low numbers of these cases as well as a presumably different etiology [50].