OUP user menu

A Meta-analysis of Alcohol Drinking and Oral and Pharyngeal Cancers: Results from Subgroup Analyses

F. Turati, W. Garavello, I. Tramacere, C. Pelucchi, C. Galeone, V. Bagnardi, G. Corrao, F. Islami, V. Fedirko, P. Boffetta, C. La Vecchia, E. Negri
DOI: http://dx.doi.org/10.1093/alcalc/ags100 107-118 First published online: 4 September 2012

Abstract

Aims: To quantify the magnitude of the association between alcohol and oral and pharyngeal cancer (OPC) by sex, smoking habits, type of alcoholic beverage and other factors. Methods: We combined findings from all case–control and cohort studies published until September 2010 and present in this article the results classified by these factors, using a meta-analytic approach. Summary relative risks (RRs) were obtained using random-effects models; heterogeneity was assessed using the χ2 test. Results: The association between alcohol and OPC risk was similar in men and women, with similar dose–response relationships. No notable differences were found with respect to geographic area and other factors, both for drinking overall and heavy (≥4 drinks/day) drinking. Among never/non-current smokers, the pooled RRs were 1.32 (95% confidence interval, CI, 1.05–1.67) for drinking, and 2.54 (95% CI, 1.80–3.58) for heavy drinking. The corresponding RRs in smokers were 2.92 (95% CI, 2.31–3.70) and 6.32 (95% CI, 5.05–7.90). The pooled RRs for any drinking irrespective of smoking were 2.12 (95% CI, 1.37–3.29) for wine-, 2.43 (95% CI, 1.92–3.07) for beer- and 2.30 (95% CI, 1.78–2.98) for spirits-only drinking. The corresponding RRs for heavy drinking were 4.92 (95% CI, 2.80–8.65), 4.20 (95% CI, 1.43–12.38) and 5.20 (95% CI, 2.77–9.78). Conclusion: The alcohol-related RRs are similar with respect to sex, geographic area and type of alcoholic beverage. The association between alcohol and OPC is stronger in smokers than in non-smokers.

INTRODUCTION

The association between alcohol consumption and oral and pharyngeal cancers (OPC) has long been established (Bagnardi et al., 2001; Brennan et al., 2008; Pelucchi et al., 2011). It has been estimated that over 30% of all cases of OPC worldwide are attributable to alcohol drinking (Boffetta et al., 2006).

In two recent meta-analyses, including >30 studies and 14,000 cases, we found a dose–response relationship between alcohol and OPC risk (relative risks, RRs, increasing from 1.29 for 10 g ethanol/day to 13.02 for 125 g ethanol/day) (Tramacere et al., 2010), with higher RRs for pharyngeal than for oral cancer, particularly at heavy doses (Turati et al., 2010).

Nevertheless, the magnitude of the association between alcohol and OPC risk by selected characteristics, such as sex, smoking habits and type of alcoholic beverage, has yet to be quantified. Thus, in order to provide estimates of such associations, we combined all published data on alcohol and OPC risk using a meta-analytic approach.

MATERIALS AND METHODS

Identification of studies and collection of data

The methodology of identification of studies and data collection has been described in a meta-analysis from our research group, published in 2010 and including 45 publications, which presented the overall results and the dose–risk relation between alcohol drinking and OPC risk (Tramacere et al., 2010). We performed a literature search using PubMed of all the original articles in English of case–control and cohort studies, using the MESH terms ‘alcohol’ and combinations of ‘mouth’ or ‘oral’ or ‘pharynx’ or ‘pharyngeal’ and ‘cancer’ or ‘carcinoma’ or ‘neoplasm’, following the Meta-analysis Of Observational Studies in Epidemiology (MOOSE) guidelines (Stroup et al., 2000).

Case–control and cohort studies considering at least three levels of alcohol consumption and reporting the odds ratio (OR) or RR or hazard ratio (HR) for oral and/or pharyngeal cancer and the corresponding confidence interval (CI)—or sufficient information to calculate them—for each exposure level were included in the present analysis.

Compared with the previously cited meta-analysis (Tramacere et al., 2010), four other publications were identified through a careful review of the reference lists in the publications retrieved (Sankaranarayanan et al., 1989; Boffetta et al., 1992; Wang et al., 2005; Allen et al., 2009), while another one (Applebaum et al., 2007) was excluded, since the results were based on the same data reported by Peters et al. (2006). Moreover, the update of the literature search to September 2010 yielded one additional publication (Shanmugham et al., 2010). Thus, we collected information on 49 publications (Wynder and Bross, 1957; Vincent and Marchetta, 1963; Keller and Terris, 1965; Martinez, 1969; Bross and Coombs, 1976; Graham et al., 1977; Elwood et al., 1984; Brugere et al., 1986; Blot et al., 1988; Tuyns et al., 1988; Merletti et al., 1989; Rossing et al., 1989; Sankaranarayanan et al., 1989; Boffetta and Garfinkel, 1990; Franceschi et al., 1990; Zheng et al., 1990; Choi and Kahyo, 1991; Oreggia et al., 1991; Boffetta et al., 1992; Kabat et al., 1994; Maier et al., 1994; Andre et al., 1995; Bundgaard et al., 1995; Takezaki et al., 1996, 2000; Hayes et al., 1999; Bouchardy et al., 2000; Moreno-Lopez et al., 2000; Garrote et al., 2001; Schwartz et al., 2001; Zavras et al., 2001; Balaram et al., 2002; Lissowska et al., 2003; Znaor et al., 2003; Altieri et al., 2004; Castellsague et al., 2004; Menvielle et al., 2004; Lee et al., 2005; Rosenquist et al., 2005; Wang et al., 2005; Peters et al., 2006; Suzuki et al., 2006; Vlajinac et al., 2006; De Stefani et al., 2007; Subapriya et al., 2007; Ide et al., 2008; Sapkota et al., 2008; Allen et al., 2009; Shanmugham et al., 2010).

In order to perform stratified analyses by sex, type of alcoholic beverage and smoking habits, 11 additional publications were considered (Wynder et al., 1957; Winn et al., 1984; Kabat and Wynder, 1989; Barra et al., 1990; Mashberg et al., 1993; Franceschi et al., 1994, 1999; De Stefani et al., 1998; Talamini et al., 1998; Fioretti et al., 1999; Bosetti et al., 2000). These provided the subgroup analyses of interest from some of the 49 studies previously identified.

For the analysis restricted to never/non-current smokers, we selected studies reporting risk estimates in subjects who never smoked in their life (Blot et al., 1988; Merletti et al., 1989; Kabat et al., 1994; Talamini et al., 1998; Fioretti et al., 1999; Hayes et al., 1999; Garrote et al., 2001; Schwartz et al., 2001; Lissowska et al., 2003; Znaor et al., 2003; Castellsague et al., 2004; Lee et al., 2005; Wang et al., 2005) and also studies considering non-current smokers (never plus ex-smokers) (Wynder et al., 1957; Bross and Coombs, 1976; Winn et al., 1984; Bundgaard et al., 1995; Subapriya et al., 2007). Some studies analyzing the association between alcohol consumption and OPC risk in light-smokers were excluded (Graham et al., 1977; Oreggia et al., 1991; Mashberg et al., 1993; Peters et al., 2006; De Stefani et al., 2007) [for the study by Merletti et al. (1989), we excluded RR estimates for men, since smokers of up to 7 g of tobacco per day]. Other seven studies, which did not publish results for alcohol drinking in never/non-current smokers in separate reports but were included in the INHANCE Consortium publication by Hashibe et al. (2007), were also considered in this subgroup analysis for the heavy drinking analysis (see Supplementary data, Appendix 1, for details). For the analysis in smokers, whenever possible, we selected RR estimates for ever smokers. When results in the original studies were presented in more than one smoking category, these categories were combined. Three publications that included in the stratified analysis by smoking habits some subjects with laryngeal cancer among the case series (Elwood et al., 1984; Tuyns et al., 1988; Peters et al., 2006), and one publication analyzing cumulative lifetime alcohol consumption across smoking strata (Zheng et al., 1990) were not considered.

For the analysis by type of alcoholic beverage, we included exclusively the studies reporting findings on consumption of wine-, beer- and/or spirits-only (detailed information on the definition of spirits consumption in the original articles is reported in the footnote of Fig. 3). For example, we included the paper published by De Stefani et al. (1998), which considered consumption of wine- and liquor-only, rather than a more updated one of the same group (De Stefani et al., 2007), reporting results from a larger dataset without distinguishing the subjects who consumed only wine or only liquor from those consuming more than one type of alcoholic beverage.

Two review team members (F.T. and I.T.) reviewed all the studies and abstracted the following information in a standard format: study design, country, number of subjects (cases, controls or cohort size), duration of follow-up (for cohort studies), years of study conduction (for case-control studies), sex of the study population, variables adjusted for in the analysis, RR estimates for categories of alcohol drinking and the corresponding 95% CI and, when available, the number of cases and non-cases for each level of alcohol consumption.

Statistical analyses

Statistical methods have been previously described (Tramacere et al., 2010). Briefly, our measure of interest was the RR or the HR for cohort and the OR for case–control studies. Whenever available, we considered multivariate risk estimates; otherwise, we utilized or computed the crude RRs (and the corresponding 95% CIs) from the distribution of cases and non-cases. When a study reported multivariate RRs but not the corresponding CIs, the SE of the adjusted estimate was obtained by penalizing the SE of the crude RR by a factor of 1.5. In the Million Women study (Allen et al., 2009), we derived the floated variances, from the 95% floated CIs provided by the authors, to obtain RRs and 95% CIs for different categories of alcohol consumption compared with non-drinkers (Easton et al., 1991).

We used gram as the measurement unit for ethanol, assuming that one drink = 12.5 g, 1 ml of ethanol = 0.8 g and 1 ounce = 28.35 g of ethanol. We assigned to each consumption category the dose corresponding to the midpoint of the range, and, for the open-ended upper category, the dose was assumed to be 1.2 times the lower bound (Berlin et al., 1993).

When possible, we chose non-drinking as the reference category. In several studies, however, occasional drinking was included in the reference category. Therefore, we defined the reference category of our analyses as ‘non- or occasional drinking’. We defined ‘moderate drinking’ as consumption of 1–2 drinks/day, and ‘heavy drinking’ as ≥4 drinks/day. When more than one study category fell in the range of alcohol consumption considered (i.e. heavy, moderate and drinking in general), we calculated the study-specific pooled estimates using, whenever possible, the method proposed by Hamling et al. (2008), taking into account the correlation between risk estimates; otherwise, we used fixed-effects models.

We derived meta-analytic summary estimates using random-effects models, which consider both within- and between-study variations (Greenland, 1987). We assessed the heterogeneity among studies using the χ² test (Greenland, 1987) (results were defined heterogeneous for P < 0.10) and the I² statistic. We performed dose–response analyses using a random-effects meta-regression model in a non-linear dose–response relationship framework, providing the best fitting two-term fractional-polynomial model (Rota et al., 2010).

For the analyses by smoking habits and type of alcoholic beverage, in addition to combining results from studies identified through the literature search, we carried out sensitivity analyses including results from the INHANCE Consortium (Hashibe et al., 2007, 2009; Purdue et al., 2009) (see Supplementary data, Appendix 1, for details).

Publication bias was evaluated through funnel plots (Thornton and Lee, 2000) and with the Begg's test (Begg and Mazumdar, 1994).

RESULTS

The characteristics of the studies have already been described (Tramacere et al., 2010). Table 1 reports the main characteristics of the five additional publications (Sankaranarayanan et al., 1989; Boffetta et al., 1992; Wang et al., 2005; Allen et al., 2009; Shanmugham et al., 2010) identified for the main analyses and not included in the previous review (Tramacere et al., 2010). In addition, information on the 11 other publications included only in the analyses stratified by sex, smoking habit or type of beverages is reported (Wynder et al., 1957; Winn et al., 1984; Kabat and Wynder, 1989; Barra et al., 1990; Mashberg et al., 1993; Franceschi et al., 1994, 1999; De Stefani et al., 1998; Talamini et al., 1998; Fioretti et al., 1999; Bosetti et al., 2000).

View this table:
Table 1.

Main characteristics of the additional five publications included in the main analyses (1) and of the 11 additional publications included in the stratified analyses only (2)

Study, yearCountryNo. of casesNo of controls/size of cohortStudy years/duration of follow-upSexVariables adjusted for in the regression modelsNotes
(1) Additional publications included in the main analyses
Case–control
Sankaranarayanan et al. (1989)India228 (158 M)453 (314 M)1983–1984Both, but RR estimates were only for MAge, religionIncluded in the main analyses
Boffetta et al. (1992)USA35922801972–1983MAge, smokingIncluded in the main analyses
Wang et al. (2005)USA3483301994–1997 (2000–2002)BothAge, smoking, alcoholIncluded in the main analyses
Cohort
Allen et al. (2009) Million Women StudyUnited Kingdom7581,280,296 PR7.2 years (average)WAge, residence, socioeconomic status, BMI, smoking, physical activity, OC, HRTIncluded in the main analyses
Shanmugham et al. (2010) Nurses' Health StudyUSA14787,621 PR1980–2006, 26 yearsWAge, follow-up time, smoking, folateIncluded in the main analyses
(2) Additional publications included in the stratified analyses only
Case–control
Wynder et al. (1957)USA543 M and 116 W207 M and 232 WNot availableBoth, but information on smoking and type of beverages was available only for MAge, religionIncluded in the analyses by smoking habits and type of alcoholic beverage only. Same study as Wynder and Bross (1957)
Winn et al. (1984)USA2274051976–1978WAge, race, residenceIncluded in the analysis by smoking habits only
Kabat and Wynder (1989)USA76115241976–1983BothAge, sex, ethnic origin, hospital, year of interviewIncluded in the drinking analysis by type of alcoholic beverage only. Same study as Kabat et al. (1994)
Barra et al. (1990)Italy30516211986–1990MWine: Age, residence, occupation, smokingIncluded in the analysis by type of alcoholic beverage only. Same study as Franceschi et al. (1990)
Beer/spirits: age, residence
Mashberg et al. (1993)USA35922801972–1983MAge, race, smoking, average alcohol consumptionIncluded in the analyses by type of alcoholic beverage and smoking habits (only for smokers stratum) only. Same study as Boffetta et al. (1992)
Franceschi et al. (1994)Italy546 (465 M)2263 (1706 M)1984–1991Both, but we considered only MAge, center, smokingIncluded in the analysis by sex (only for men; women were already included in the study as Bosetti et al. (2000)) only. Same study as Franceschi et al. (1990)
De Stefani et al. (1998)Uruguay4714711988–1996MAge, residence, urban/rural status, smokingIncluded in the analysis by type of alcoholic beverage only. Same study as De Stefani et al. (2007)
Talamini et al. (1998)Italy and Switzerland60 (never smokers)692 (never smokers)1992–1997BothAge, sex, center, educationIncluded in the analysis by smoking habits (only for never/non-current smokers stratum) only. Same study as Altieri et al. (2004)
Fioretti et al. (1999)Italy42 (never smokers)864 (never smokers)1984–1993BothAge, center, sex, educationIncluded in the analysis by smoking habits (only for never/non-current smokers stratum) only. Same study as Franceschi et al. (Franceschi et al., 1990)
Franceschi et al. (1999)Italy and Switzerland63812541992–1997MAge, residence, interviewer, education, smoking, vegetable and fruit intake, total energy intakeIncluded in the analysis by sex and smoking habits (only for smokers stratum) only. Same study as Altieri et al. (2004)
Bosetti et al. (2000)Italy and Switzerland19511131984–1997WAge, center, education, BMI, smokingIncluded in the analysis by sex only. Same study as Altieri et al. (2004) and Franceschi et al. (1990)
  • RR, relative risk; PR, persons at risk; PY, person-year; M, men; W, women; BMI, body mass index; OC, oral contraceptives; HRT, hormone replacement therapy.

Data on alcohol drinking were available from 49 studies (18,387 cases), while the analyses on moderate and heavy alcohol drinking were based on 33 and 31 studies, respectively. There was no evidence of publication bias overall (P for Begg's test for the drinking analysis = 0.127; P for Begg's test for the heavy drinking analysis = 0.708).

By study design, sex, geographic region and other covariates

Drinking

Table 2 reports summary statistics for the association between alcohol drinking and OPC risk in strata of study design, sex, geographic area and other covariates. Compared with non- or occasional drinking, the overall RR for drinking was 2.55 (95% CI, 2.15–3.02) and the corresponding estimates for case–control and cohort studies were 2.70 (95% CI, 2.31–3.16) and 1.34 (95% CI, 0.88–2.05), respectively, with significant heterogeneity between study designs (P < 0.001). The pooled RRs were 3.12 (95% CI, 2.50–3.89) for men and 1.52 (95% CI, 1.17–1.98) for women, with significant heterogeneity between estimates (P < 0.001). When the analysis was restricted to studies reporting RR estimates for both men and women separately (i.e. studies carried out on males or females only were excluded), the summary RRs were 2.33 (95% CI, 1.65–3.29) for men and 1.84 (95% CI, 1.34–2.54) for women (P for heterogeneity = 0.325) (for this specific purpose, we considered the paper by Franceschi et al., (1994) which reported ORs for men and women, rather than the one by Bosetti et al. (2000), which provided results from the same study but on women only). No differences in RR estimates were found across strata of geographic area, type of outcome (incidence or mortality) and smoking adjustment.

View this table:
Table 2.

Summary statistics for the association between alcohol drinking and OPC risk in strata of selected covariates

 Drinking vs. non- or occasional drinking
No. of studiesRandom-effects RR (95% CI)P-values for heterogeneity testI2 statistics (%)
All studies492.55 (2.15–3.02)<0.00192.2
Study design
 Case–control452.70 (2.31–3.16)<0.00187.9
 Cohort41.34 (0.88–2.05)<0.00191.0
P heterogeneity between strata < 0.001
Sexa
 Men293.12 (2.50–3.89)<0.00190.7
  Case–control studies273.22 (2.54–4.08)<0.00191.2
  Cohort studies22.17 (1.71–2.74)0.8740.0
 Women131.52 (1.17–1.98)<0.00174.2
  Case–control studies101.82 (1.42–2.34)0.11237.1
  Cohort studies31.01 (0.91–1.12)0.8330.0
P heterogeneity between strata < 0.001
Geographic area
 Asia112.28 (1.84–2.83)<0.00193.6
 Other than Asia382.61 (2.11–3.23)<0.00193.6
  North America152.31 (1.89–2.83)<0.00181.6
  South America52.59 (1.85–3.62)0.16238.9
  Europe182.87 (1.91–4.30)<0.00196.5
P heterogeneity between stratab = 0.369
P heterogeneity between stratac = 0.730
Outcome
 Incidence472.58 (2.16–3.08)<0.00192.5
 Mortality22.11 (1.68–2.66)0.4810.0
P heterogeneity between strata = 0.175
Smoking adjustment
 No72.61 (2.00–3.40)0.02060.0
 Yes422.52 (2.08–3.05)<0.00193.1
P heterogeneity between strata = 0.833
Reference category
 Non/occasional drinking382.18 (1.88–2.52)<0.00185.2
 Different from non/occasional drinkingd114.11 (2.74–6.15)<0.00192.7
P heterogeneity between strata = 0.004
Smoking habits
 Never/non-current smokers191.32 (1.05–1.67)0.20720.3
  Never smokers141.34 (1.02–1.77)0.10433.9
  Non-current smokers51.24 (0.73–2.11)0.5870.0
 Smokers202.92 (2.31–3.70)<0.00177.7
  Ever smokers123.76 (2.80–5.05)<0.00181.4
  Current smokers81.94 (1.43–2.62)0.10840.5
P heterogeneity between stratae < 0.001
P heterogeneity between strataf < 0.001
  • OPC, oral and pharyngeal cancers; RR, relative risk; CI, confidence interval.

  • aTwenty studies gave information on men only (Wynder and Bross, 1957; Keller and Terris, 1965; Graham et al., 1977; Brugere et al., 1986; Tuyns et al., 1988; Sankaranarayanan et al., 1989; Boffetta and Garfinkel, 1990; Zheng et al., 1990; Oreggia et al., 1991; Boffetta et al., 1992; Franceschi et al., 1994, 1999; Maier et al., 1994; Andre et al., 1995; Takezaki et al., 2000; Balaram et al., 2002; Znaor et al., 2003; Menvielle et al., 2004; Lee et al., 2005; De Stefani et al., 2007) and four on women only (Bross and Coombs, 1976; Bosetti et al., 2000; Allen et al., 2009; Shanmugham et al., 2010); nine studies presented risk estimates separately for sex (Vincent and Marchetta, 1963; Martinez, 1969; Blot et al., 1988; Merletti et al., 1989; Choi and Kahyo, 1991; Kabat et al., 1994; Hayes et al., 1999; Zavras et al., 2001; Ide et al., 2008).

  • bTest for heterogeneity between pooled risk estimates from ‘Asia’ and ‘Other than Asia’.

  • cTest for heterogeneity between pooled risk estimates from ‘Asia’, ‘North America’, ‘South America’ and ‘Europe’.

  • dThis category includes studies in which the reference category included the consumption of amount of alcohol >0.5 drinks/day.

  • eTest for heterogeneity between pooled risk estimates from ‘never/non-current smokers’ and ‘smokers’.

  • fTest for heterogeneity between pooled risk estimates from ‘never smokers’ and ‘ever smokers’.

Moderate drinking

The pooled RR for moderate drinking (i.e. 1–2 drinks/day) was 1.36 (95% CI, 1.20–1.54); the corresponding estimates for case–control and cohort studies were 1.44 (95% CI, 1.30–1.60) and 0.90 (95% CI, 0.80–1.00), respectively (data not shown). The overall RR for men was 1.28 (95% CI, 1.08–1.51) and the one for women was 1.17 (95% CI, 0.92–1.49). Compared with non- or occasional drinking, the overall RRs were 1.58 (95% CI, 1.23–2.03) for 11 Asian studies and 1.28 (95% CI, 1.11–1.48) for 22 non-Asian ones. When the analysis was restricted to studies with smoking adjustment, the summary RRs for moderate drinking were 1.62 (95% CI, 1.26–2.08) for 10 Asian studies and 1.27 (95% CI, 1.08–1.50) for 17 non-Asian ones (P for heterogeneity = 0.111) (Fig. 1). No notable differences emerged with respect to type of outcome and reference category (data not shown).

Fig. 1.
Fig. 1.

Forest plot for the association between moderate alcohol drinking (1–2 drinks/day) and OPC risk in strata of geographic region. Only smoking-adjusted studies were included in the meta-analysis. Pooled RR = 1.21 (95% CI, 0.98–1.48) for eight North American studies; RR = 1.20 (95% CI, 0.92–1.56) for four South American studies; RR = 1.48 (95% CI, 1.01–2.17) for five European studies.

Heavy drinking

Table 3 shows summary statistics for the association between heavy alcohol drinking and OPC risk in strata of selected covariates. Compared with non- or occasional drinking, the overall RRs for heavy drinking were 5.40 (95% CI, 4.49–6.50) for all the 31 studies combined, 5.51 (95% CI, 4.54–6.69) for case–control and 4.25 (95% CI, 3.03–5.96) for cohort studies (P for heterogeneity = 0.402). No significant heterogeneity in risk estimates was found with respect to sex (see also Supplementary data, Appendix 2), geographic area and other covariates.

View this table:
Table 3.

Summary statistics for the association between heavy alcohol drinking (≥4 drinks/day) and OPC risk in strata of selected covariates

 Heavy drinking vs. non- or occasional drinking
No. of studiesRandom-effects RR (95% CI)P-values for heterogeneity testI2 statistics (%)
All studies315.40 (4.49–6.50)<0.00178.1%
Study design
 Case–control295.51 (4.54–6.69)<0.00179.0%
 Cohort24.25 (3.03–5.96)0.5510.0%
P heterogeneity between strata = 0.402
Sexa
 Men225.49 (4.36–6.92)<0.00179.7%
  Case–control studies205.67 (4.43–7.26)<0.00180.9%
  Cohort studies24.25 (3.03–5.96)0.5510.0%
 Women35.69 (3.74–8.66)0.3950.0%
P heterogeneity between strata = 0.878
Geographic area
 Asia24.75 (3.14–7.17)0.3900.0%
 Other than Asia295.73 (4.51–6.64)<0.00179.3%
  North America105.36 (4.11–7.00)<0.00176.6%
  South America55.21 (3.77–7.19)0.29618.7%
  Europe145.63 (4.09–7.77)<0.00182.1%
P heterogeneity between stratab = 0.420
P heterogeneity between stratac = 0.935
Outcome
 Incidence315.51 (4.54–6.69)<0.00179.0%
 Mortality24.25 (3.03–5.96)0.5510.0%
P heterogeneity between strata = 0.402
Smoking adjustment
 No36.53 (3.04–14.03)0.00186.2%
 Yes285.33 (4.39–6.46)<0.00177.7%
P heterogeneity between strata = 0.614
Reference category
 Non/occasional drinking215.22 (4.38–6.22)0.00253.0%
 Different from non/occasional drinking105.75 (4.49–8.60)<0.00190.1%
P heterogeneity between strata = 771
Smoking habits
 Never/non-current smokersd162.54 (1.80–3.58)0.26216.7%
  Never smokersd142.53 (1.73–3.72)0.15727.8%
  Non-current smokers22.63 (0.55–12.72)0.9870.0%
 Smokers116.32 (5.05–7.90)0.04545.6%
  Ever smokers86.67 (5.30–8.38)0.08743.7%
  Current smokers34.82 (2.43–9.59)0.11254.3%
P heterogeneity between stratae <0.001
P heterogeneity between strataf <0.001
  • OPC, oral and pharyngeal cancers; RR, relative risk; CI, confidence interval.

  • aInformation on heavy drinking was available from 22 of the 31 studies providing RR estimates for men and from 3 of the 13 studies providing RR estimates for women.

  • bTest for heterogeneity between pooled risk estimates from ‘Asia’ and ‘Other than Asia’.

  • cTest for heterogeneity between pooled risk estimates from ‘Asia’, ‘North America’, ‘South America’ and ‘Europe’.

  • dIncluding seven studies from the INHANCE Consortium [i.e. (1) Central Europe Study, (2) Seattle Study, (3) North Carolina Study, (4) Tampa Study, (5) Los Angeles Study, (6) Huston Study and (7) International multicenter Study] whose results were reported in Fig. 1 of the publication by Hashibe et al. (2007).

  • eTest for heterogeneity between pooled risk estimates from ‘never/non-current smokers’ and ‘smokers’.

  • fTest for heterogeneity between pooled risk estimates from ‘never smokers’ and ‘ever smokers’.

Dose–response analysis

Supplementary data, Appendix 3, shows the best-fitting dose–response curve describing the relation between alcohol drinking and OPC risk in men and women [i.e. ln(RR) = (dose) + (dose)2 in men; ln(RR) = (dose)2 + (dose)3 in women]. These functions show RRs of 1.35 (95% CI, 1.25–1.45) for 12 g, 1.84 (95% CI, 1.60–2.12) for 25 g, 3.23 (95% CI, 2.53–4.12) for 50 g, 5.39 (95% CI, 3.93–7.39) for 75 g and 8.55 (95% CI, 5.87–12.45) for 100 g of ethanol/day in men. In women, the dose–response analysis estimates RRs of 1.09 (95% CI, 1.05–1.13) for 12 g, 1.40 (95% CI, 1.25–1.58) for 25 g and 3.31 (95% CI, 2.65–4.15) for 50 g of ethanol/day (data on heavier alcohol doses were not available from the original studies).

By smoking habits

Drinking

Compared with non- or occasional drinking, the pooled RRs for drinking were 1.32 (95% CI, 1.05–1.67) for never/non-current smokers (RR = 1.34 for never and RR = 1.24 for non-current smokers) and 2.92 (95% CI, 2.31–3.70) for smokers (RR = 3.76 for ever and RR = 1.94 for current smokers), with heterogeneity between these risk estimates (P < 0.001) (Table 2 and Panel 1 of Fig. 2). When results from the INHANCE Consortium were included in the meta-analysis (Hashibe et al., 2007, 2009), the pooled RRs for drinking were 1.37 (95% CI, 1.14–1.63) for never/non-current smokers and 2.45 (95% CI, 2.05–2.94) for smokers.

Fig. 2.

Forest plots for the association between drinking (Panel 1) and heavy (≥4 drinks/day) (Panel 2) drinking and OPC risk in never smokers, non-current smokers, ever smokers and current smokers subjects. Panel 2: In the never smokers subgroup, RRs were for ≥3 drinks/day in the seven INHANCE studies (Hashibe et al., 2007), >21 drinks/week for the study by Wang et al. (2005) and >40 g/day for the study by Merletti et al. (1989). When these risk estimates were excluded from the meta-analysis, since referring to an amount of alcohol not reaching the established cut-off for heavy drinking (i.e. 4 drinks/week), the pooled RR for never/non-current smokers combined was 2.79 (95% CI, 1.80–4.33). In the ever smokers subgroup, RR was for >21 drinks/week for the study by Wang et al. (2005). When this risk estimate was excluded from the meta-analysis, the pooled RR for ever and current smokers combined was 6.26 (95% CI, 4.89–8.01).

Heavy drinking

Compared with non- or occasional drinking, the overall RRs for heavy drinking were 2.54 (95% CI, 1.80–3.58) for never/non-current smokers (RR = 2.53 for never and RR = 2.63 for non-current smokers) and 6.32 (95% CI, 5.05–7.90) for smokers (RR = 6.67 for ever and RR = 4.82 for current smokers) (Table 3 and Panel 2 of Fig. 2), with significant heterogeneity between risk estimates (P < 0.001). Exclusion of studies whose RRs referred to an amount of alcohol not reaching the established cut-off for heavy drinking (i.e. 4 drinks/day) did not materially change our results. When results from the pooled analysis by the INHANCE Consortium (Hashibe et al., 2007, 2009) were considered, the overall RRs for heavy drinking were 2.79 (95% CI, 1.80–4.33) for never/non-current smokers and 5.91 (95% CI, 4.47–7.82) for smokers.

By type of alcoholic beverage

Drinking

Figure 3 shows forest plots for the association between OPC risk and wine-, beer- and spirits-only drinking (Panel 1), and for heavy (≥4 drinks/day) wine- and beer-only drinking (Panel 2) (only one study reported information on heavy spirits-only drinking). Compared with non- or occasional drinking, the pooled RRs for drinking were 2.12 (95% CI, 1.37–3.29) for wine-, 2.43 (95% CI, 1.92–3.07) for beer- and 2.30 (95% CI, 1.78–2.98) for spirits-only. No heterogeneity was found between types of beverages (P = 0.856). When results from the INHANCE Consortium (Purdue et al., 2009) were included [and results from the study by Altieri et al. (2004) and Barra et al. (1990) were excluded], the corresponding summary RRs were 1.67 (95% CI, 1.20–2.33) for wine-, 2.56 (95% CI, 2.18–3.01) for beer- and 2.38 (95% CI, 1.93–2.93) for spirits-only consumption (data not shown).

Fig. 3.

Forest plots for the association between OPC risk and wine-, beer- and spirits-only drinking (Panel 1). Forest plots for heavy (≥4 drinks/day) wine-only and beer-only drinking (Panel 2). Panel 1: Spirits consumption was defined as consumption of whiskey in two publications (Keller and Terris, 1965; Mashberg et al., 1993), hard liquor in one publication (Kabat and Wynder, 1989; De Stefani et al., 1998), spirits in two publications (Barra et al., 1990; Zheng et al., 1990), Ouzo and/or Tsipouro in one publication (Zavras et al., 2001) and arrack-country liquor-spirits in one publication (Znaor et al., 2003). Panel 2: When the study by Altieri et al. (2004), defining heavy wine drinking as ≥3 drinks/day, was excluded, the overall RR was 3.72 (95% CI, 2.63–5.26).

Heavy drinking

Compared with non- or occasional drinking, the pooled RRs for heavy drinking were 4.92 (95% CI, 2.80–8.65) for wine-, 4.20 (95% CI, 1.43–12.38) for beer- and 5.20 (95% CI, 2.77–9.78) for spirits-only (Fig. 3, Panel 2). No heterogeneity was found between RRs for wine, beer and spirits (P = 0.945).

DISCUSSION

This meta-analysis including 49 publications and ∼18,000 OPC cases worldwide provided quantitative evidence for the relation between alcohol and OPC risk by several subgroups of interest, including smoking status and type of alcoholic beverage, using a systematic meta-analytic approach.

We found associations of similar magnitude between alcohol and OPC risk in men and women; although the best fitting models were mathematically different, they led to fairly similar dose–risk relations. The weaker association observed in women for alcohol drinking in general may be attributed to the higher average amount of alcohol consumption in men compared with women.

Results of this meta-analysis also suggest that alcohol consumption increases OPC risk even in the absence of tobacco use. However, the association between alcohol and OPC risk is significantly weaker in never/non-current smokers than in smokers, especially at heavy doses, suggesting that alcohol may interact with smoking in further increasing OPC risk, in more than a multiplicative way (Franceschi et al., 1990). The interaction between alcohol and tobacco has been examined by several studies. Most of them showed that subjects who are heavily exposed to both risk factors are subject to greater than additive risks, which exceed 30-fold in many studies (Blot et al., 1988; Merletti et al., 1989; Zheng et al., 1990; Oreggia et al., 1991; Mashberg et al., 1993; Kabat et al., 1994; Andre et al., 1995; Bundgaard et al., 1995; Franceschi et al., 1999; Hayes et al., 1999; Garrote et al., 2001; Castellsague et al., 2004; Lee et al., 2005; De Stefani et al., 2007; Weikert et al., 2009). Moreover, a recent publication from the INHANCE Consortium found that the joint effect of tobacco smoking and alcohol drinking is more than expected under the multiplicative model for OPC (Hashibe et al., 2009).

It has been suggested that the most frequently consumed alcoholic beverage in a given culture is associated with the highest risk at equivalent levels of consumption (e.g. wine in Southern Europe, beer and hard liquor in North America, sake in Japan and cachaca in Brazil) (Franceschi et al., 1990; Mayne et al., 2006). This may be explained by the greater overall alcohol consumption in users of the most common type of alcohol. Moreover, information on intake is collected probably more thoroughly for the most frequently consumed alcoholic beverage than for the other beverages in various populations. When we combined RRs associated with wine-, beer- or spirits-only consumption from countries characterized by different alcohol drinking pattern, we found similar pooled RRs for each alcoholic beverage, for both drinkers of any amount and heavy drinkers. We were not able to analyze moderate consumption (i.e. 1–2 drinks/day), given the limited available data. In the INHANCE Consortium, a much weaker association with moderate consumption of wine-only than of beer-only and spirits-only was observed. The authors, however, stated that they cannot exclude that those findings are attributable to residual confounding from diet and other lifestyle factors (Purdue et al., 2009). Moreover, the analysis by geographic region showed similar associations between alcohol in general and OPC risk in populations coming from different areas. These observations are in agreement with the hypothesis that various types of alcoholic beverages are carcinogenic, and support the role of ethanol and its metabolites, mainly acetaldehyde, as principal carcinogenic agents rather than beverage-specific compounds (Boccia et al., 2009; Seitz and Stickel, 2010).

The association between any and heavy alcohol drinking with OPC cancer was consistent across several subgroups. However, we found higher pooled risk estimates in case–control than cohort studies, which are more sensitive to problems of recall and/or selection bias (Breslow and Day, 1980). It is also possible, however, that information of cohort studies had been collected long before diagnosis and consequently reflects drinking habits in the distant past.

Limitations of our meta-analysis include selection bias and confounding in observational studies, the limited data from cohort studies (particularly for the heavy drinking analyses and for the analyses by smoking habit and type of alcoholic beverage), the criticisms related to retrospective exposure assessment in case–control studies and the possible residual confounding by tobacco or other risk factors. It is possible that alcohol consumption is systematically underreported by both cases and controls in several studies. This would in most, although in not all, instances lead to an overestimation of the RR, particularly for moderate and light doses (Breslow and Day, 1980; Doll et al., 1994; Bagnardi et al., 2001). However, studies investigating reproducibility and validity of self-reported alcohol drinking in various populations found satisfactory correlation coefficients (Giovannucci et al., 1991; Ferraroni et al., 1996; Flagg et al., 2000; Horn-Ross et al., 2008). A limitation of the dose–response analysis is that it assumes a dose–response with no threshold. Since the slope of the function depends on the level of misclassification in the different categories of alcohol consumption, if heavy drinking is more frequently misclassified than drinking at lower doses, the slope of the dose-risk function at low doses will be over-estimated.

The major strengths of our meta-analysis were the collection of a uniquely large number of OPC cases, which enabled us to explore in detail the association of interest among selected subgroups, and the use of a systematic meta-analytic approach to summarize our results.

In conclusion, the alcohol-related RRs are similar with respect to sex, geographic area and type of alcoholic beverage. The association between alcohol and OPC is stronger in smokers than in never/non-current smokers, suggesting an over multiplicative effect of alcohol and tobacco on oral and pharyngeal carcinogenesis.

Funding

This work was supported by the Italian Association for Cancer Research (AIRC) (projects No. 10068 and No. 10258, My First AIRC Grant), and the Flight Attendants Medical Research Institute Center of Excellence (Award 052460_CoE). C.G. was supported by Fondazione Veronesi. I.T. was supported by a fellowship from the Italian Foundation for Cancer Research (FIRC). The study sponsors had no role in the conduct of the analysis and writing of the manuscript. The authors thank Ms I. Garimoldi for editorial assistance.

Conflict of interest statement. None declared.

References

View Abstract