Alcohol consumption and lung cancer risk in never smokers: a meta-analysis

Alcohol consumption and lung cancer risk in never smokers: a meta-analysis

original article Annals of Oncology 22: 2631–2639, 2011 doi:10.1093/annonc/mdr027 Published online 22 March 2011 Alcohol consumption and lung cancer...

335KB Sizes 0 Downloads 31 Views

original article

Annals of Oncology 22: 2631–2639, 2011 doi:10.1093/annonc/mdr027 Published online 22 March 2011

Alcohol consumption and lung cancer risk in never smokers: a meta-analysis V. Bagnardi1,2*, M. Rota1,3, E. Botteri2,4, L. Scotti1, M. Jenab5, R. Bellocco1,6, I. Tramacere7, C. Pelucchi7, E. Negri7, C. La Vecchia4,7, G. Corrao1 & P. Boffetta8,9

Background: The role of alcohol consumption as an independent risk factor for lung cancer is controversial. Since drinking and smoking are strongly associated, residual confounding by smoking may bias the estimation of alcohol consumption and lung cancer risk relation. Therefore, we undertook a meta-analysis to quantitatively assess the association between alcohol and risk of lung cancer in never smokers. Methods: After a literature search in Medline, we included all case–control and cohort studies published up to January 2010 that reported an estimate of the association between alcohol intake and lung cancer risk in never smokers. Results: We selected 10 articles, including 1913 never smoker lung cancer cases. The random-effects pooled relative risk (RR) for drinkers versus nondrinkers was 1.21 [95% confidence interval (CI) 0.95–1.55]. The same figure was 1.05 (95% CI 0.89–1.23) after the exclusion of one outlier study. At the dose–response analysis, RR for an increase in alcohol intake of 10 g/day was 1.01 (95% CI 0.92–1.10). Conclusions: Alcohol consumption was not associated with lung cancer risk in never smokers. Even if the synergistic effect of smoking and alcohol cannot be ruled out, our results suggest that alcohol does not play an independent role in lung cancer etiology. Key words: alcohol, lung cancer, meta-analysis, never smokers

introduction The role of alcohol consumption on the occurrence of lung cancer is still controversial. Since drinking is strongly related to smoking, the positive association between alcohol and lung cancer reported in several studies [1–4] is possibly due to the residual confounding by smoking. A simple yet effective way to clarify whether alcohol is an independent risk factor for lung cancer is to limit the investigation to never smokers; while a pooled analysis based on seven large cohort studies suggested that alcohol consumption might play an independent role in the etiology of lung cancer in male never smokers [5], other prospective studies resulted in no association [6, 7] or a protective effect [8]. Since limiting the investigation of lung cancer to the subgroup of never smokers leads to small sample size and fewer subjects exposed to alcohol [9], these conflicting results could be due to random error. Therefore, to enhance power to identify the potential association between alcohol drinking and lung cancer risk in never smokers, *Correspondence to: Dr V. Bagnardi, Division of Epidemiology and Biostatistics, European Institute of Oncology, Via Ripamonti 435, 20141 Milan, Italy. Tel: +39-0257489820; Fax: +39-02-57489922; E-mail: [email protected]

we conducted a meta-analysis of published epidemiological data investigating this issue.

methods search strategy and study selection We carried out a Medline search (from 1960 to January 2010) for studies investigating the association between alcohol consumption and lung cancer risk. The search strategy is described in Box 1. In addition, the reference lists of retrieved articles and of reviews and meta-analyses published on the issue were hand-checked to identify additional relevant studies. We limited our search to studies published in English. An article was included in the meta-analysis only if it satisfied the following criteria: 1 case–control or cohort studies published as original articles (abstracts, letters, reviews and meta-analyses were excluded); 2 studies that reported findings expressed as odds ratio (OR), relative risk (RR) or hazard ratio (or reported sufficient data to compute them) in never smokers, with alcohol intake considered as an exposure and lung cancer incidence or mortality as an outcome; and 3 studies that reported standard errors or confidence intervals (CIs) of the risk estimates or provided sufficient data to calculate them.

ª The Author 2011. Published by Oxford University Press on behalf of the European Society for Medical Oncology. All rights reserved. For permissions, please email: [email protected]

original article

Received 1 December 2010; revised 13 January 2011; accepted 17 January 2011

Downloaded from http://annonc.oxfordjournals.org/ at Robertson Library University of prince Edward Island on July 13, 2015

1 Department of Statistics, University of Milan-Bicocca, Milan; 2Division of Epidemiology and Biostatistics, European Institute of Oncology, Milan; 3Department of Clinical Medicine and Prevention, Centre of Biostatistics for Clinical Epidemiology, University of Milan-Bicocca, Monza; 4Institute of Medical Statistics and Biometrics ‘G. A. Maccacaro’, University of Milan, Milan, Italy; 5International Agency for Research on Cancer, Lyon, France; 6Department of Medical Epidemiology and Biostatistics, Karolinska Institute, Stockholm, Sweden; 7Mario Negri Institute for Pharmacological Research, Milan, Italy; 8The Tisch Cancer Institute, Mount Sinai School of Medicine, New York, USA; 9International Prevention Research Institute, Lyon, France

original article Box 1: Search strategy and terms used to identify studies on alcohol consumption and lung cancer risk

In addition, studies that met the above criteria but reported only information on specific type of alcoholic beverages (i.e. beer, wine or spirits) without reporting on total alcohol consumption were excluded because nondrinkers of a specific beverage might drink other types of alcoholic beverages. Three investigators (VB, LS and MR) independently determined the eligibility of each article for inclusion in the meta-analysis. Any disagreement was resolved by consensus among the three. When the results of the same study were published in more than one paper, only the most recent and/or complete article was included in the analysis.

data extraction For each included study, we extracted details on study design, country, gender, categories of alcohol intake considered, RR estimates and 95% CIs, adjustment variables and, when available, the number of cases and controls (case–control studies) or the number of events and subjects at risk (cohort studies) for the reported exposure levels. Since lung cancer is a relatively rare event, we assumed that ORs, risk ratios and rate ratios were all comparable estimates of the RR [10]. Where possible, separate estimates were extracted for males and females.

data analysis A pooled RR of lung cancer for drinkers versus nondrinkers was computed, using both fixed-effects and, in case of significant heterogeneity between estimates, random-effects models [11]. When a study reported adjusted risk estimates for two or more categories of exposure compared with a common baseline group, we aggregated them in a single estimate relative to drinkers versus nondrinkers category using the method for pooling nonindependent estimates described by Hamling et al. [12]. This method uses the dose-specific covariate-adjusted risk estimates, their CIs and the crude numbers of cases and non-cases for each category of exposure in order to generate a set of pseudo-numbers of cases and non-cases consistent with both the adjusted estimates and their CIs. The pseudo-numbers of two or more categories of exposure can then be combined to originate adjusted risk estimates for any new risk category. In our case, we combined all the categories for drinkers in a single category and compared it with the reference nondrinkers category. Statistical heterogeneity among studies was evaluated using both the Q statistics and I2, which is the proportion of total variation contributed by between-study variance [13].

2632 | Bagnardi et al.

2413 papers identified through MEDLINE search and screened by titles and abstracts

2306 papers excluded because they were judged not to be related to alcohol drinking and lung cancer in humans or because they were not published in English

107 full text papers retrieved for detailed assessment

2 additional papers identified through review of reference lists of retrieved papers

99 papers excluded because they did not satisfied the following inclusion criteria: 25 were not case-control or cohort studies published as original articles; 45 did not reported effect estimates of alcohol on lung cancer risk; 5 reported only information on specific types of alcoholic beverage; 5 reported data published in other papers; 19 did not reported effect estimates in never smokers

10 papers included in the meta-analysis on alcohol consumption and lung cancer risk in never smokers

Figure 1. Selection strategy flow diagram.

An influence analysis was conducted and the summary RR was computed with the omission of one study at a time, in order to identify to what extent the results were influenced by single studies. In order to assess the effect of study characteristics and quality on the reported effect estimates, we conducted heterogeneity analyses in which we compared the summary effect estimates for subgroups stratified on study or quality characteristics such as study design (cohort or case–control), geographic area (North America, Europe or Asia), definition of ‘never smokers’ used in the study (never smoked regularly, <100 cigarettes lifetime or unspecified), gender (males, females or males and females) and adjustment for potential confounders (variables related to diet, body mass index, socioeconomic status or educational level). We were not able to investigate adjustment for physical activity and environmental tobacco smoking, given the sparse number of studies that considered these factors. We used a chi-square statistics to test for differences of summary estimates among subgroups [10]. Publication bias was evaluated using Egger’s regression asymmetry test [14]. We then assessed the dose–response relationship between alcohol intake and lung cancer risk in never smokers. In this analysis, we considered only studies reporting risk estimates for at least three quantitative exposure categories, including the referent one. Data extraction from the original studies for the dose–response metaanalysis was carried out according to a three-step procedure. First, since different studies used different units of measure to express alcohol consumption (grams, milliliters, ounces or drinks consumed every day, week, month or year), we used grams per day (g/day) as a standard measure of ethanol intake using the following equivalencies: 0.8 g/ml, 28 g/ounce, 12.5 g/drink. Secondly, since the levels of consumption were often given by a range, the value x of exposure was assigned as the midpoints of the ranges of the reported categories of alcohol intake (as suggested by Berlin et al. [15], the x values were calculated as 1.2 times the lower bound for the openended upper category). Thirdly, every measure of association concerning

Volume 22 | No. 12 | December 2011

Downloaded from http://annonc.oxfordjournals.org/ at Robertson Library University of prince Edward Island on July 13, 2015

MeSH terms search 1. ‘ethanol’; 2. ‘alcohol drinking’; 3. ‘lung neoplasm’; 4. [(1) or (2)] and 3. Direct keyword search 5. alcohol; 6. alcoholic beverages; 7. ethanol; 8. alcohol drinking; 9. lung neoplasm; 10. lung cancer; 11. [(5) or (6) or (7) or (8)] and [(9) or (10)]; 12. (4) or (11).

Annals of Oncology

original article

Annals of Oncology

Table 1. Characteristics of studies included in the meta-analysis on alcohol consumption and lung cancer among never smokers Study design

First author, publication year (reference), country

Definition of never smokers

Sex

No. of cases

Cohort

Thun, 2009 [7], USA

Never smoked regularly

M

406

Case–control (population)

Cohort

Cohort

Bagnardi, 2010 [4], Italy

Benedetti, 2006 [19], Canada

Nishino, 2006 [18], Japan

Rohrmann, 2006 [6], Europe

<100 cigarettes lifetime

Never smoked regularly

Unspecified

Unspecified

Volume 22 | No. 12 | December 2011

72 969

W

625

150 247

M/W

125

637

W

M

M/W

33

18

97

308

6033

143 893

Alcohol intake categories Intake at recruitment (drink/day) 0 <1 1 2–3 4+ 0 <1 1 2–3 4+ Intake during adulthood (g/day)

0 0.1–4.9 5–14.9 15–29.9 30–59.9 ‡60 Intake during drinking periods (drink/week) Never weekly 1–6 ‡7 Intake at recruitment (g/day)

Reported RR (95% CI)

Controlled variables

Age, occupation, education, ethnicity 1.00 0.99 1.04 0.99 0.84 1.00 1.14 1.04 1.00 0.69

(referent) (0.78–1.26) (0.71–1.53) (0.75–1.31) (0.55–1.27) (referent) (0.95–1.35) (0.76–1.43) (0.77–1.30) (0.41–1.16) Age, sex, education, study area, BMI, passive smoking, fruit and vegetable consumption, fresh red and processed meat consumption

1.06 1.00 0.79 1.17 0.55 1.41

(0.54–2.07) (referent) (0.41–1.52) (0.65–2.13) (0.25–1.21) (0.15–12.94) Age, ethnicity, SES, education

1.00 (referent) 0.2 (0.0–0.6) 1.1 (0.4–3.3) Age, familial history of lung cancer, green leafy vegetables consumption, orange and fruit consumption

Never drinkers <24.9 25–49.9 ‡50 Ex-drinkers Ever drinkers Intake at recruitment (g/day)

1.00 1.10 0.37 1.15 4.20 1.22

(referent) (0.35–3.51) (0.04–3.18) (0.13–9.98) (1.12–15.72) (0.43–3.45)

0 0.1–4.9 5–14.9

0.64 (0.33–1.23) 1.00 (referent) 0.93 (0.56–1.53)

Age, sex, study area, height, weight, diet, education, physical activity, fruit consumption, fresh red and processed meat consumption

doi:10.1093/annonc/mdr027 | 2633

Downloaded from http://annonc.oxfordjournals.org/ at Robertson Library University of prince Edward Island on July 13, 2015

Case–control (population)

No. at risk/controls

original article

Annals of Oncology

Table 1. (Continued) Study design

Pooled analysis of cohort studies

Case–control (population)

Case–control (hospital)

Case–control (population)

Freudenheim, 2005 [5], USA, Canada and The Netherlands

Hu, 2002 [20], Canada

Rachtan, 2002 [21], Poland

Zang, 2001 [22], USA

Koo, 1988 [23], China

Definition of never smokers

Unspecified

Unspecified

Unspecified

Never smoked regularly

<20 cigarettes lifetime

Sex

M

No. of cases

74

No. at risk/controls

59 882

W

183

168 561

W

157

465

W

M

W

54

53

88

251

2091

137

Alcohol intake categories

Reported RR (95% CI)

15–29.9 ‡30 Intake at recruitment (g/day)

0.67 (0.32–1.41) 0.55 (0.17–1.83)

0 0.1 to <5 5 to <15 ‡15 0 0.1 to <5 5 to <15 ‡15 Intake at 2 years before interview (drink/week) 0 1 >1 Usual consumption (g/week) <1 1–4 4–8 ‡8 Lifelong intake (oz/day whiskey equivalenta) <1 1–5.9 ‡6 Intake at 1 year before interview (drink/week) <1 ‡1

1.00 1.49 2.53 6.38 1.00 0.98 0.89 1.35

Controlled variables

Education, BMI, energy intake

(referent) (0.64–3.49) (1.10–5.81) (2.74–14.90) (referent) (0.69–1.37) (0.52–1.54) (0.64–2.87) Age, province of residence, education, social class

1.00 (referent) 0.8 (0.5–1.4) 0.8 (0.5–1.2) Age

1.00 3.89 8.76 12.06

(referent) (1.82–8.32) (2.81–27.29) (3.94–36.91) Age, BMI

1.00 (referent) 1.2 (0.7–2.1) 0.7 (0.2–2.0) Age, no. of live births, education 1.00 (referent) 1.85b

a

1 oz = 11.5 g of alcohol. CI not reported. RR, relative risk; CI, confidence interval; M, men; W, women; BMI, body mass index; SES, socioeconomic status.

b

each level of alcohol consumption and the corresponding CI was translated into log(RR) and the corresponding variance. The method proposed by Greenland and Longnecker [16] was used to compute the linear trend from the correlated log RRs across categories of alcohol consumption. This approach is based on constructing an approximate covariance estimate for the log(RRs) and estimating a linear trend using generalized least-squares regression. We considered zero exposure dose (nondrinkers) as reference. When a study reported risk estimates and 95% CIs relative to a referent category different from zero, we recalculated the corresponding RRs using nondrinkers as reference. The variance of the logarithm of these new

2634 | Bagnardi et al.

estimates was obtained using the approximate covariance of the original log(RRs), according to the formula proposed by Greenland et al. [17]. Heterogeneity among study-specific trend estimates and subgroup analyses were carried out. All analyses were carried out with SAS software, version 9.1 (SAS Institute Inc., Cary, NC). All P values were two sided.

results Figure 1 shows the paper selection strategy for the metaanalysis.

Volume 22 | No. 12 | December 2011

Downloaded from http://annonc.oxfordjournals.org/ at Robertson Library University of prince Edward Island on July 13, 2015

Case–control (population)

First author, publication year (reference), country

original article

Annals of Oncology

Table 2. Pooled RRs of lung cancer for drinkers versus nondrinkers by study subgroups Category

Study characteristics (number of studies)

Summary estimate

95% CI

I2 (%)

Study design

Case–control (6) Cohort (4) North America (4) Europe (3) Asia (2) North America and Europe (1) Never smoked regularly (3) <100 cigarettes lifetime (2) Unspecified (5) Male (4)a Female (6)a Male and female (2)a Yes (4) No (6) Yes (3) No (7) Yes (7) No (3) (10)

1.25 1.02 0.98 1.89 1.64 1.50 1.01 1.23 1.60 1.22 1.26 1.07 1.20 1.22 1.14 1.26 1.05 1.99 1.21

(0.68–2.31) (0.92–1.28) (0.85–1.13) (0.62–5.77) (0.97–2.78) (0.58–3.90) (0.88–1.17) (0.57–2.65) (0.90–2.85) (0.83–1.80) (0.81–1.95) (0.64–1.80) (0.84–1.72) (0.87–1.71) (0.78–1.66) (0.91–1.74) (0.87–1.26) (0.63–6.32) (0.95–1.55)

86 31 20 90 0 – 13 71 85 51 86 0 44 83 53 81 51 88 74

Geographic area

Definition of never smokers

Gender

Adjustment for diet factors Adjustment for BMI Adjustment for SES and/or educational level Overall

P value for heterogeneity among subgroups 0.664

0.160

0.291

0.882 0.954 0.699 0.283 –

a

Sum does not add up to the total number of studies included in the meta-analysis since studies reporting gender-specific estimates were considered as independent strata. RR, relative risk; CI, confidence interval; BMI, body mass index; SES, socio-economic status.

We identified via the search strategy a total of 2413 papers. We screened them by title and abstract and excluded 2306 of them because they were not related to alcohol

Volume 22 | No. 12 | December 2011

intake and lung cancer risk or because they were not published in English-language journals. The remaining 107 articles were considered of interest and full text was retrieved for detailed

doi:10.1093/annonc/mdr027 | 2635

Downloaded from http://annonc.oxfordjournals.org/ at Robertson Library University of prince Edward Island on July 13, 2015

Figure 2. RRs of lung cancer for drinkers versus nondrinkers. Squares indicate study-specific RRs. Size of the square is proportional to the precision of the estimate (i.e. the inverse of the variance). Horizontal lines indicate the 95% CI). Diamond indicates pooled RR with its corresponding 95% CI. RR, relative risk; CI, confidence interval.

original article

drinkers versus nondrinkers Figure 2 shows the RR of lung cancer in never smokers for drinkers versus nondrinkers in each of the 10 included studies.

Only 2 studies [18, 23] reported directly this aggregate estimate; for the remaining 10 studies, it was derived combining the reported dose-specific RRs. We found a substantial heterogeneity among single studies estimates (Q = 42.85, P < 0.0001, I2 = 74%). The random-effect summary RR was 1.21 (95% CI 0.95–1.55). Influence analysis showed that the heterogeneity was due in a large part to one study [21], reporting elevated ORs for lung cancer among never smoker women associated with very low levels of alcohol intake (ORs equal to 3.89, 8.76 and 12.06 for 1–4, 4–8 and ‡8 g/week, respectively). When omitting this study, the I2 dropped to 39%, and the summary RR became 1.05 (95% CI 0.89–1.23). We examined possible differences between risk estimates by various study characteristics (Table 2). We did not find evidence of heterogeneity in pooled estimates by design, gender, definition of ‘never smokers’ and area in which the study was carried out. When considering adjustment for potential confounders, we did not find a significant difference between estimates adjusted and those not adjusted for diet factors, body mass index and socioeconomic status and/or educational level.

Figure 3. Study-specific RR estimates of lung cancer for increasing level of alcohol consumption. Each panel refers to a study. The first author, year of publication and gender (in square brackets) of subjects included in the analysis are reported. Black squares indicate the RR estimates and whiskers their 95% CIs. Continuous lines describe the log-linear relationship between RR and alcohol dose (g/day). Since the regression line is forced to pass through the origin (i.e. RR = 1), it does not have to interpolate the reported study-specific estimates. The vertical axis is on a log scale. RR, relative risk; CI, confidence interval.

2636 | Bagnardi et al.

Volume 22 | No. 12 | December 2011

Downloaded from http://annonc.oxfordjournals.org/ at Robertson Library University of prince Edward Island on July 13, 2015

evaluation. Two other additional papers were identified through review of reference lists of retrieved papers. Eighty of these 109 articles were subsequently excluded from the metaanalysis because they did not satisfy the inclusion criteria. A further 19 epidemiological studies investigating the association between alcohol consumption and lung cancer risk were excluded since they did not report any risk estimate in never smokers. Three cohort studies [6, 7, 18]; 1 pooled analysis of 7 prospective studies conducted in the United States, Canada and The Netherlands [5]; and 6 case–control studies [4,19–23], leading to a total of 10 studies, met the inclusion criteria and were included in the analysis. Overall, the meta-analysis was based on a total of 1913 lung cancer cases. The main characteristics of the included studies are reported in Table 1.

Annals of Oncology

original article

Annals of Oncology

dose–risk Two of the 10 studies were excluded from the dose–response meta-analysis. The first one [23] because it considered only one level of exposure. The second one [21] because it reported, as previously described, a very narrow range of exposure levels associated with elevated risk estimates: in fact, according to the data reported, we obtained an implausible log-linear regression coefficient for an increase of 1 g/day (1.97) leading to an RR of lung cancer for an increase of 10 g/day of 3.83 · 108. Figure 3 shows the estimated study-specific log-linear risk function for increasing level of alcohol consumption, along with the RR (and 95% CI) of each study. There was significant heterogeneity among study-specific slopes (Q = 27.1, P = 0.0013, I2 = 67%). The random-effects model-based summary RR of lung cancer for an increase in alcohol intake of 10 g/day was 1.01 (95% CI 0.92–1.10). We examined possible differences between linear trend risk estimates by various study characteristics (Table 3). No evidence of heterogeneity between subgroup-specific pooled estimates was found.

discussion In this meta-analysis of published data, we observed that lung cancer risk is not associated with alcohol drinking in never smokers. No positive trend in lung cancer risk was observed.

Table 3. Pooled RRs of lung cancer for an increase in alcohol intake of 10 g/day by study subgroups Category

Study characteristics (number of studies)

Summary estimate

95% CI

I2 (%)

Study design

Case–control (4) Cohort (4) North America (4) Europe (2) Asia (1) North America and European Union (1) Never smoked regularly (3) <100 cigarettes lifetime (1) Unspecified (4) Male (4)a Female (4)a Male and female (2)a Yes (4) No (4) Yes (3) No (5) Yes (6) No (2) (8)

0.97 1.02 0.98 0.93 0.95 1.68

(0.89–1.06) (0.92–1.19) (0.94–1.02) (0.82–1.04) (0.68–1.33) (0.70–4.04)

0 80 0 0 – –

0.98 0.93 1.17 1.15 0.97 0.93 1.12 0.98 1.20 0.98 1.02 0.99 1.01

(0.94–1.02) (0.81–1.07) (0.76–1.79) (0.93–1.44) (0.92–1.03) (0.82–1.04) (0.86–1.47) (0.94–1.02) (0.91–1.60) (0.94–1.01) (0.92–1.14) (0.89–1.11) (0.92–1.10)

0 – 83 87 0 0 80 0 87 0 74 0 67

Geographic area

Definition of never smokers

Gender

Adjustment for diet factors Adjustment for BMI Adjustment for SES and/or educational level Overall

P value for heterogeneity among subgroups 0.303

0.507

0.552

0.221 0.325 0.150 0.690 –

a

Sum does not add up to the total number of studies included in the meta-analysis since studies reporting gender-specific estimates were considered as independent strata. RR, relative risk; CI, confidence interval; BMI, body mass index; SES, socio-economic status.

Volume 22 | No. 12 | December 2011

doi:10.1093/annonc/mdr027 | 2637

Downloaded from http://annonc.oxfordjournals.org/ at Robertson Library University of prince Edward Island on July 13, 2015

Previous meta-analyses showed a modest increase in smoking-adjusted risk limited to very high consumption of alcohol [24, 25]. Since residual confounding by smoking is one of the main concerns in the interpretation of risk among heavy drinkers [26], we chose to restrict the analysis only to those studies reporting the association between alcohol and lung cancer in never smokers. Founded on this restriction, the strongest evidence in favor of an independent role of alcohol consumption in lung cancer etiology derives from a large pooled analysis [5] in which a sixfold significant increase of risk was reported in men who drank ‡15 g/day but did not smoke compared with men who did not drink or smoke. On the other hand, the same study reported no association whatsoever in nonsmoking women. In contrast, another cohort study [7] reported no evidence of association in 72 969 never smoker males, where 406 deaths from lung cancer were observed during 24 years of follow-up, as well as in 150 247 never smoker females, where 625 deaths were observed. In our meta-analysis, the drinkers versus nondrinkers and dose–response pooled RRs were not significantly different between men and women. Although a substantial heterogeneity was found across single studies estimates, this could not be explained by gender, study designs, geographic areas nor by adjustment for diet and socioeconomic factors. The heterogeneity could be possibly due to differences in smoking and alcohol drinking assessment among studies and different drinking patterns among study populations. Two studies by Shimazu et al. [8] and Murata et al. [27] are in support of the lack of a positive association between alcohol

The asymmetry test for publication bias was not statistically significant (P = 0.29).

original article

2638 | Bagnardi et al.

smokers, suggesting that alcohol consumption does not likely play an independent role from tobacco smoking in lung cancer etiology.

acknowledgement The funders had no role in the study design, data collection and analysis, decision to publish or preparation of the manuscript.

funding Italian Association for Cancer Research (MFAG-10258); University of Milan-Bicocca (FAR-12-1-5151630-239); Flight Attendants Medical Research Institute Center of Excellence (Award 052460_CoE).

disclosure The authors declare no conflict of interest.

references 1. De Stefani E, Correa P, Fierro L et al. The effect of alcohol on the risk of lung cancer in Uruguay. Cancer Epidemiol Biomarkers Prev 1993; 2(1): 21–26. 2. Dosemeci M, Gokmen I, Unsal M et al. Tobacco, alcohol use, and risks of laryngeal and lung cancer by subsite and histologic type in Turkey. Cancer Causes Control 1997; 8(5): 729–737. 3. Prescott E, Gronbaek M, Becker U, Sorensen TI. Alcohol intake and the risk of lung cancer: influence of type of alcoholic beverage. Am J Epidemiol 1999; 149(5): 463–470. 4. Bagnardi V, Randi G, Lubin J et al. Alcohol consumption and lung cancer risk in the Environment and Genetics in Lung Cancer Etiology (EAGLE) study. Am J Epidemiol 2010; 171(1): 36–44. 5. Freudenheim JL, Ritz J, Smith-Warner SA et al. Alcohol consumption and risk of lung cancer: a pooled analysis of cohort studies. Am J Clin Nutr 2005; 82(3): 657–667. 6. Rohrmann S, Linseisen J, Boshuizen HC et al. Ethanol intake and risk of lung cancer in the European Prospective Investigation into Cancer and Nutrition (EPIC). Am J Epidemiol 2006; 164(11): 1103–1114. 7. Thun MJ, Hannan LM, DeLancey JO. Alcohol consumption not associated with lung cancer mortality in lifelong nonsmokers. Cancer Epidemiol Biomarkers Prev 2009; 18(8): 2269–2272. 8. Shimazu T, Inoue M, Sasazuki S et al. Alcohol and risk of lung cancer among Japanese men: data from a large-scale population-based cohort study, the JPHC study. Cancer Causes Control 2008; 19(10): 1095–1102. 9. Blair A, Freeman LB. Lung cancer among nonsmokers. Epidemiology 2006; 17(6): 601–603. 10. Greenland S. Quantitative methods in the review of epidemiologic literature. Epidemiol Rev 1987; 9: 1–30. 11. DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials 1986; 7(3): 177–188. 12. Hamling J, Lee P, Weitkunat R, Ambuhl M. Facilitating meta-analyses by deriving relative effect and precision estimates for alternative comparisons from a set of estimates presented by exposure level or disease category. Stat Med 2008; 27(7): 954–970. 13. Higgins JP, Thompson SG. Quantifying heterogeneity in a meta-analysis. Stat Med 2002; 21(11): 1539–1558. 14. Egger M, Davey SG, Schneider M, Minder C. Bias in meta-analysis detected by a simple, graphical test. BMJ 1997; 315(7109): 629–634. 15. Berlin JA, Longnecker MP, Greenland S. Meta-analysis of epidemiologic doseresponse data. Epidemiology 1993; 4(3): 218–228. 16. Greenland S, Longnecker MP. Methods for trend estimation from summarized dose-response data, with applications to meta-analysis. Am J Epidemiol 1992; 135(11): 1301–1309.

Volume 22 | No. 12 | December 2011

Downloaded from http://annonc.oxfordjournals.org/ at Robertson Library University of prince Edward Island on July 13, 2015

and lung cancer also in the broader category of nonsmokers, which includes never smokers as well as ex-smokers. The first reported a protective effect of alcohol on lung cancer risk; the latter reported a lack of association. When we added these two studies to our meta-analysis, we obtained a pooled RR of lung cancer for an increase in alcohol intake of 10 g/day of 1.00. Some limitations characterize our study. First, 1913 patients could be considered too small a population to support our conclusions. We observed that only 10 of the 29 studies investigating the association between alcohol and lung cancer reported estimates in never smokers strata. However, we think that many authors did not report data in never smokers simply because of the null results and/or the small number of never smokers. Secondly, the effect of specific alcoholic beverages was not investigated. Studies evaluating the effect of alcohol on lung cancer by beverage type have more frequently observed relations with beer and liquor than with wine [25]. Unfortunately, only two studies [20, 21] reported beveragespecific estimates in never smokers, but with conflicting results. Thirdly, passive smoking may act as a confounder in the observed associations since nonsmoking heavy drinkers may have indeed more chance to be exposed to passive smoking than nonsmoking nondrinkers. Unfortunately, only 1 study [4] of the 10 controlled the estimates for passive smoking, showing no effect of alcohol on lung cancer risk. Finally, misclassification of smoking status could represent another issue in our meta-analysis [9]. However, results from validation studies conducted within case–control studies on lung cancer risk factors generally reported low proportions of misclassified ever-regular smokers among reported never smokers [28]. The fact that the results of this meta-analysis, limited to never smokers, do not support an independent role of alcohol on lung cancer risk does not imply that alcohol does not play a role in the etiology of lung cancer in general. For instance, we cannot conclude that alcohol is not associated with lung cancer in smokers. A synergic effect between alcohol drinking and tobacco smoking on lung cancer risk was recently reported in a large case–control study [4]. Alcohol may enhance the carcinogenic effect of cigarette smoke on lung tissues by inducing the activity of cytochrome P-450 enzymes, which in turn can activate procarcinogens present in alcoholic beverages [29]. Also, all the studies included in our meta-analysis considered lung cancer as if it were a homogeneous disease. However, lung cancer is a collection of several conditions, with different histologies, anatomical subsites and genetic profiles. Therefore, our null results do not exclude the independent role of alcohol on a particular subtype of lung cancer. The association between smoking and lung cancer, for example, varies among histological types, being stronger with small-cell lung cancer and squamous cell carcinoma and weaker with adenocarcinoma [30]. Lung cancer in never smokers ranks as the seventh most common cause of cancer death worldwide and elucidating the etiology of this condition is important from a public health prospective [31]. This meta-analysis, based on 1913 never smoker cases, provides no evidence that alcohol consumption is positively associated with lung cancer risk at any dose in never

Annals of Oncology

Annals of Oncology

Volume 22 | No. 12 | December 2011

24. Bagnardi V, Blangiardo M, La Vecchia C, Corrao G. A meta-analysis of alcohol drinking and cancer risk. Br J Cancer 2001; 85(11): 1700–1705. 25. Chao C. Associations between beer, wine, and liquor consumption and lung cancer risk: a meta-analysis. Cancer Epidemiol Biomarkers Prev 2007; 16(11): 2436–2447. 26. Boffetta P. Alcohol and lung cancer: do we have the answers? Am J Clin Nutr 2005; 82(3): 495–496. 27. Murata M, Takayama K, Choi BC, Pak AW. A nested case-control study on alcohol drinking, tobacco smoking, and cancer. Cancer Detect Prev 1996; 20(6): 557–565. 28. Nyberg F, Agudo A, Boffetta P et al. European validation study of smoking and environmental tobacco smoke exposure in nonsmoking lung cancer cases and controls. Cancer Causes Control 1998; 9(2): 173–182. 29. Poschl G, Seitz HK. Alcohol and cancer. Alcohol 2004; 39(3): 155–165. 30. Khuder SA. Effect of cigarette smoking on major histological types of lung cancer: a meta-analysis. Lung Cancer 2001; 31(2–3): 139–148. 31. Sun S, Schiller JH, Gazdar AF. Lung cancer in never smokers—a different disease. Nat Rev Cancer 2007; 7(10): 778–790.

doi:10.1093/annonc/mdr027 | 2639

Downloaded from http://annonc.oxfordjournals.org/ at Robertson Library University of prince Edward Island on July 13, 2015

17. Greenland S, Michels KB, Robins JM et al. Presenting statistical uncertainty in trends and dose-response relations. Am J Epidemiol 1999; 149(12): 1077–1086. 18. Nishino Y, Wakai K, Kondo T et al. Alcohol consumption and lung cancer mortality in Japanese men: results from Japan collaborative cohort (JACC) study. J Epidemiol 2006; 16(2): 49–56. 19. Benedetti A, Parent ME, Siemiatycki J. Consumption of alcoholic beverages and risk of lung cancer: results from two case-control studies in Montreal, Canada. Cancer Causes Control 2006; 17(4): 469–480. 20. Hu J, Mao Y, Dryer D, White K. Risk factors for lung cancer among Canadian women who have never smoked. Cancer Detect Prev 2002; 26(2): 129–138. 21. Rachtan J. Alcoholic beverages consumption and lung cancer cell types among women in Poland. Lung Cancer 2002; 35(2): 119–127. 22. Zang EA, Wynder EL. Reevaluation of the confounding effect of cigarette smoking on the relationship between alcohol use and lung cancer risk, with larynx cancer used as a positive control. Prev Med 2001; 32(4): 359–370. 23. Koo LC. Dietary habits and lung cancer risk among Chinese females in Hong Kong who never smoked. Nutr Cancer 1988; 11(3): 155–172.

original article