Quantification of the association between alcohol drinking and risk of esophageal squamous cell carcinoma (ESCC) is an open issue, particularly among light alcohol drinkers, never-smokers, and Asian populations, in which some high-risk polymorphisms in alcohol metabolizing genes are more prevalent. To address these issues, we conducted a systematic review and meta-analysis using 40 case-control and 13 cohort studies that reported on the risk associated with alcohol drinking for at least three levels of consumption. In studies adjusted for age, sex, and tobacco smoking, the relative risk (RR) and 95% confidence interval (CI) for the association between light alcohol drinking (≤12.5 g/d) and risk of ESCC was 1.38 (1.14–1.67). The association was slightly stronger in Asian countries than in other populations. The adjusted RRs (95% CIs) were 2.62 (2.07–3.31) for moderate drinking (>12.5–<50 g/d) and 5.54 (3.92–7.28) for high alcohol intake (≥50 g/d); the RRs were slightly higher in non-Asian populations. In prospective studies, the RR (95% CI) was 1.35 (0.92–1.98) for light, 2.15 (1.55–2.98) for moderate, and 3.35 (2.06–5.46) for high alcohol intakes; light drinking showed an association with ESCC in Asia (five studies) but not in other regions (three studies). Among never-smokers (nine studies), the RR (95% CI) was 0.74 (0.47–1.16) for light, 1.54 (1.09–2.17) for moderate, and 3.09 (1.75–5.46) for high intakes. This meta-analysis further corroborates the association of moderate and high alcohol intake with risk of ESCC and provides risk estimates based on multiple prospective studies. Light alcohol intake appears to be associated to ESCC mainly in studies in Asia, which suggests a possible role of genetic susceptibility factors.
An association between alcohol drinking and risk of esophageal cancer has long been reported. In 1987, and again in 2009, within the Monographs on the Evaluation of Carcinogenic Risks to Humans, the International Agency for Research on Cancer (IARC) working group classified alcoholic beverages as “carcinogenic to humans” and concluded that esophageal cancers are causally related to the consumption of alcoholic beverages.1, 2 The alcohol related increased risk of esophageal cancer has mainly been observed with esophageal squamous cell carcinoma (ESCC).3 In the past decades, the incidence rate of esophageal adenocarcinoma has increased in several Western countries, while the incidence rate of ESCC has been either stable or decreased in those countries.4 However, ESCC is still the most common histological type of esophageal cancer worldwide, particularly in high-incidence areas.4
The dose-risk relation between alcohol drinking and risk of esophageal cancer has been investigated in a few meta-analyses.5–7 While the association of alcohol drinking in general and heavy alcohol drinking with esophageal cancer has been well-established in previous analyses, the magnitude of the association with light alcohol drinking needs further investigation, because only a limited number of studies had reported on light exposure when those meta-analyses were conducted. Furthermore, the earlier analyses were mainly based on case-control studies, as very few prospective studies were available at that time. In the past few years, several prospective studies on the risk of esophageal cancer associated with several levels of alcohol consumption have been published. Tobacco use is another major risk factor for ESCC, which has shown a multiplicative effect with alcohol drinking.1 To account for the confounding effects of smoking, most studies on alcohol drinking and ESCC have adjusted their results for tobacco use. However, the potential for residual confounding remains even after adjustments. Investigation of the association between alcohol intake and risk of ESCC among never-smokers usually showed difficult, because of paucity of alcohol drinker cases who did not smoke. However, several studies that have reported the association among never-smokers are now available, but they have not yet been included in any meta-analysis. A regional difference in risk is also possible, because of a potential different pattern of alcohol metabolism in Southeast Asia.
To address these points, we conducted an updated systematic review and meta-analysis to investigate patterns of the association between alcohol drinking and risk of ESCC using all relevant case-control and prospective studies, with additional focus on patterns with light alcohol drinking, in prospective studies, among never-smokers, and by geographic area.
Material and Methods
Selection of studies
Our inclusion criteria were: (1) observational studies; (2) studies with data on total alcohol intake, excluding studies reporting specific alcoholic beverage but not total alcohol consumption, because nondrinkers of a specific beverage might drink other types of alcoholic beverages; (3) studies in which conversion of alcohol consumption to daily intake of ethanol was possible, for example, excluding studies reporting duration of intake only; (4) studies reporting estimates of relative risks (RR), that is, any measure of effect including odds ratio, rate ratio, and hazard ratio, and the corresponding confidence intervals (CI)—or sufficient information to calculate them—at least for three levels of alcohol consumption (reference group plus 2 other groups); and (5) studies reporting on ESCC or all esophageal cancer combined, excluding the studies reporting on several cancer sites combined, for example, upper aerodigestive tract cancers. We excluded studies/data on esophageal adenocarcinoma or other histologic subtypes when the results were reported by histologic subtypes. We assumed that the majority of cases in studies from non-Western countries were ESCC.4 In Western countries, the majority of recent studies reported their results by histologic subtype. As the rise in the incidence of adenocarcinoma in those countries mainly occurred in the most recent decades,4 we assumed that the majority of cases in the earlier studies that did not report on histologic subtype had ESCC.
As a starting point, we used a list of publications prepared for an earlier meta-analysis on alcohol drinking and risk of several cancers. The list covered relevant articles published up to 2000.5 To update this list, we searched the PubMed database for all case-control or cohort studies published in English from January 1, 1999 to June 2, 2010 on the association between alcohol consumption and risk of ESCC using the following terms: “Esophageal Neoplasms” [MeSH Terms] and [cohort OR prospective OR (case-control) OR (case control)]. Using this approach, 2,494 publications were retrieved. The abstract of all articles and then the full text of 171 potentially eligible publications were examined. We also searched the bibliographies of relevant original and review articles and systematic reports.1, 2, 7 We identified 127 eligible publications using our starting list and results of the PubMed and reference list search.
When both center-based and pooled results were available for multicenter studies, we only included the pooled ones. We also included the most informative article when data from an individual study were available in several articles. After excluding articles with repeated data, publications from 53 studies were included in this meta-analysis.3, 8–59 Two of the articles had partial overlap with a few other studies.20, 36 Of the 53 publications, 5 studies also provided data on alcohol drinking among never-smokers.3, 23, 46, 52, 54 There were four additional articles that reported on alcohol drinking among never-smokers,60–63 while the results on their entire study population (smokers and nonsmokers combined) were presented in other included articles.10, 19, 29, 30 Data from those three articles were only used in our analyses for never-smokers, for which a total of nine publications were finally included.
Data extraction and statistical analysis
Where data were available, we extracted information on study design, study area and period, number of participants, inclusion of men and women in the study, the source of controls for case-control studies, and the variables for which study results were controlled (by matched analyses, stratified analyses, or adjustments in multivariate analyses), as well as the number of cases and noncases, RRs, and 95% CIs for each alcohol consumption level. We used the maximally adjusted results when both crude and adjusted RRs and 95% CIs were presented. When RRs were not presented, unadjusted estimates were computed using the exposure distributions that were given in the articles. Two studies reported adjusted RRs without CIs;10, 23 to calculate adjusted CIs, we considered the standard error of the adjusted RR as equal to the standard error of the unadjusted RR multiplied by 1.5.
Alcohol consumption in two articles was reported by equivalent volumes to an alcoholic beverage with 25% content of ethanol18 or by hard liquor intake volumes22; we divided the consumption values by 4 and 2.5, respectively, to calculate the volume of absolute alcohol intake. All other studies reported the quantity of alcohol consumption as volume/weight of ethanol or the number of drinks/units using defined alcohol content per drink/unit. We extracted the amount of daily consumption and converted it to grams of ethanol per day using the following equivalencies: 1 drink = 12.5 g (unless stated otherwise in the article), 1 ml of pure alcohol = 0.8 g, and 1 ounce of pure alcohol = 28.35 g. We assigned the midpoint of each exposure category as the dose associated to each RR estimate. For open-ended upper categories, we multiplied the lower bound of the categories by 1.2 and, to examine the impact of the choice of this factor, alternatively by 1.5. As the results in both instances were similar, we only report those obtained using the first value (1.2-times). The reference category in the majority of articles consisted of nondrinkers. In some studies, occasional drinkers were also included in the reference category. In two studies,49, 56 nondrinkers were classified as a separate category, but the risk estimates were based on a reference group other than nondrinkers, that is, the drinkers that had the lowest intakes compared with other drinkers. For those studies, we combined nondrinkers and the reference groups and estimated crude RRs (95% CIs) for different levels of alcohol consumption using exposure distributions.
Daily amount of alcohol consumption was categorized as no/occasional intake, light (≤12.5 g ethanol; ≤1 drink), moderate (>12.5 to <50 g ethanol; >1 to < 4 drinks), and high (≥50 g ethanol; ≥ 4 drinks). The studies generally used different cut-off points for alcohol consumption levels. To obtain risk estimates for our defined exposure levels, we sometimes pooled more than one estimate using the method proposed by Hamling et al.64 This method uses the number of exposed and nonexposed subjects and the original risk estimates to derive a set of pseudo-numbers of cases and controls/subjects at risk by taking into account the correlation between the original estimates. These pseudo-numbers can be used to calculate adjusted risk estimates and CIs for any alternative exposure level; we pooled the pseudo-cases and pseudo-controls according to our selected cut-off points.
Summary RRs and 95% CIs were derived using random-effects models. We conducted several subgroup analyses, including: (1) only prospective studies; (2) only studies with more precise estimates, defined as those with standard error < 0.5 in logarithmic scale for light and moderate intake; as our analyses showed publication bias with articles reporting on high alcohol intake, we included studies with standard error <0.3 in this subgroup analysis to also eliminate the influence of publication bias; (3) only studies with population-based controls, as the pattern of alcohol drinking among hospital controls with certain diseases and in the general population might be different; (4) only estimates exclusively for ESCC (not combined with other histological subtypes); (5) by geographic area (Asia versus other regions); (6) only studies that presented RRs adjusted for main potential confounding factors (age, sex, and tobacco use); (7) by sex; (8) excluding the two studies that reported adjusted RRs without 95% CIs;10, 23 (9) excluding the two studies with potential partial overlap with a few other studies;20, 36 (10) only studies with exclusively nondrinkers as the reference category (excluding occasional-drinkers)3, 8, 9, 11, 17–19, 22, 23, 26, 27, 29, 31–33, 35–37, 39, 40, 42, 44–46, 48, 50, 52–55, 58, 59; and (11) by outcome (mortality12, 15, 17, 47, 51, 59 versus incidence).
Heterogeneity among studies was estimated using the I2 statistic. To examine publication bias, we plotted Begg's funnel plot and used Egger's weighted regression method. All statistical analyses were performed using SAS (SAS Institute Inc., NC; version 9.2) and Stata (StataCorp LP, TX; version 11) statistical software.
Summary characteristics of the 40 retrospective case-control and 13 prospective studies included in this meta-analysis are presented in Supporting Information Tables 1 and 2, respectively. Seventeen studies were from America, 22 from Asia, 1 from Australia, and 13 from Europe.
Summary statistics for the association between alcohol drinking and ESCC are shown in Table 1. Data were available for light intake from 26 studies, for moderate intake from 47 studies, and for high intake from 39 studies. The forest plots for the association between alcohol drinking and risk of ESCC by drinking level are shown in Figures 1–3. The RR (95% CI) for light alcohol drinking was 1.31 (1.10–1.57) in overall analysis and 1.35 (0.92–1.98) for prospective studies only. With moderate alcohol drinking, the RR (95% CI) was 2.27 (1.89–2.72) for the overall analysis and 2.15 (1.55–2.98) for prospective studies only. Heavy alcohol drinking was strongly associated with the risk of ESCC in overall analysis (RR = 4.89; 95% CI, 3.84–6.23) and somewhat less pronounced in prospective studies only (RR = 3.35; 95% CI, 2.06–5.46). The Begg's funnel plots by drinking level are shown in Supporting Information Figure 1. Although the Egger's p-value was marginally significant for light alcohol drinking (p = 0.03), the funnel plot was fairly symmetric and did not suggest major publication bias. The funnel plot and Egger's weighted regression method (p for bias = 0.10) did not suggest publication bias for moderate alcohol drinking. However, both the funnel plot and Egger's method (p for bias = 0.004) suggest significant publication bias for heavy alcohol drinking. Results of subgroup analyses for studies with more precise estimates, studies with population-based controls, data on ESCC only, and adjusted results were qualitatively similar to the results of overall analysis or prospective studies only. However, with moderate and high intake, the estimates were slightly higher for the ESCC only subgroup and adjusted results. I2 statistics suggested high heterogeneity for all above analyses, with p-values ≤ 0.002 (Table 1).
Table 1. Summary statistics for the association between alcohol consumption and esophageal squamous cell carcinoma
The analyses by geographic region (Table 2) showed a significant association between light alcohol drinking and risk of ESCC in Asia in all analyses, including overall analysis (RR = 1.63; 95% CI: 1.20–2.22), adjusted studies (RR = 1.52; 95% CI: 1.06–2.19), and prospective studies (RR = 1.89; 95% CI: 1.49–2.41). The latter association had the lowest heterogeneity among the analyses by geographic area (I2 statistics = 17.8%, p = 0.30). In the other regions, the RR (95% CI) was 1.17 (0.99–1.39) for the overall analysis, 1.28 (1.04–1.59) for adjusted studies, and 1.03 (0.76–1.39) for prospective studies. Point estimates for moderate alcohol intake were slightly higher in non-Asian populations. For high alcohol intake, the RR was higher in non-Asian countries in overall analysis and adjusted studies, but it was smaller in those countries in prospective studies; however, there were only three studies in the latter group, with heterogeneous results.3, 17, 56
Table 2. Summary statistics for the association between alcohol consumption and esophageal squamous cell carcinoma by study region1
We do not report results of the other subgroup analyses, for example, by sex, because there were only three studies reporting heavy drinking in women and the other estimates were similar to the above results. Furthermore, none of the analyses substantially explained the observed heterogeneity among the studies.
Summary characteristics of the nine studies reporting on the association between alcohol drinking and risk of ESCC among never-smokers are indicated in Supporting Information Tables 1 and 2. The overall RR (95% CI; the number of studies) were 0.74 (0.47–1.16; n = 5) for light intake, 1.54 (1.09–2.17; n = 9) for moderate intake, and 3.09 (1.75–5.46; n = 8) for high intake (Fig. 4). The p-values for heterogeneity were 0.27 for light intake, 0.17 for moderate intake and 0.02 for high intake. There were too few studies to investigate the association in prospective studies and in Asian countries. The funnel plot (Supporting Information Figure 2) and Egger's weighted regression method (p for bias = 0.44) did not suggest publication bias.
Moderate and high alcohol drinking were associated with an increased risk of ESCC in both Asian and non-Asian countries and in all other subgroup analyses. With regard to light alcohol drinking results were more complex, showing an increased risk of ESCC mainly in studies in Asian countries. Results of analyses among never-smokers also showed associations of moderate and high drinking with ESCC.
The association of moderate and heavy alcohol drinking with risk of ESCC was robust across several subgroups. The risk estimates were somewhat higher in case-control studies than in prospective studies. This may be partly related to potential bias associated with uncertainties in retrospective assessment of exposure in case-control studies (recall bias). Furthermore, there was significant publication bias for the studies reporting high alcohol intake. When we examined studies with more precise estimates in that group (standard error < 0.3), which did not show any publication bias, the results were very similar to those of prospective studies, indicating that the publication bias was mainly originated from case-control rather than prospective studies. Therefore, RRs from prospective studies seem to give more valid risk estimates for high drinking group. The magnitude of associations among never-smokers was similar to those in prospective studies, implying that recall bias or residual confounding of tobacco use might also have influenced the associations with alcohol consumption observed in case-control studies. The majority of studies that reported adjusted results and histological subtypes were case-control studies, which tended to show slightly higher RRs than overall estimates, but the difference was not substantial. The above finding for ESCC cases only supports our assumption when we were selecting relevant publications: ESCC probably constituted the majority of cases in the studies that had not reported histology, since it seems that there is little or no association between alcohol drinking and risk of esophageal adenocarcinoma.3, 48, 54, 56 ESCC has a poor prognosis.7 There was no major difference in risk estimates between studies reporting on incidence and on mortality.
The association between light drinking and ESCC in Asian populations was robust across all subgroups. Whereas overall and adjusted studies showed a significant association between light drinking and ESCC also in non-Asian countries, such an association was not observed in prospective studies from those countries and among never-smokers. The above groups, however, included only three prospective studies and five studies with small numbers of never-smoker cases and controls, respectively. Consequently, there is no robust evidence for an association between light drinking and ESCC in non-Asian countries, but such an association cannot be ruled out either.
The stronger association between light drinking and risk of ESCC in Asian populations may be related to effect modification by genetic factors. Acetaldehyde is the first metabolite of ethanol, and it has been related to esophageal cancer.2 Aldehyde dehydrogenase 2 (ALDH2) is the major enzyme that eliminates acetaldehyde.1 Several studies have investigated the association between genes involved in metabolism of alcohol/acetaldehyde and risk of ESCC, and polymorphisms in CYP1A1 and ALDH2 have shown consistent associations.65 Polymorphism in ALDH2 leads to ∼4-fold and 13-fold higher serum acetaldehyde levels among heterozygotes and polymorphic homozygotes, respectively.65 However, polymorphic homozygotes usually experience flushing and other unpleasant symptoms such as nausea and headache following alcohol drinking, even with light intake.38 Heterozygotes may tolerate alcohol intake to some degree, but they can have the symptoms with higher intakes.38 Therefore, while heterozygotes are at ∼3-fold higher risk of ESCC, individuals with homozygote polymorphism are at lower risk (approximately one-third) compared to wild type homozygotes, because polymorphic homozygotes usually abstain from alcohol drinking.65 Polymorphic ALDH2 is prevalent in East Asia (up to 40%), while it is rare in Western populations (<5%).66 In this meta-analysis, all of the included studies from Asia were from populations with higher prevalence of polymorphic ALDH2 compared to Western populations.
A few studies have investigated the interaction between alcohol drinking and polymorphism in ALDH2 in relation to risk of ESCC. Although they generally showed a significant interaction between heavy alcohol drinking and the polymorphism, they presented little robust information on light drinking, perhaps because of high variation in definition of light drinking and generally small number of participants in that category of alcohol intake.67 Our findings support the potential role of genetic susceptibility in the risk associated with light drinking in Asian countries. The observed pattern of risk of ESCC in Asian and other population, that is, a similar pattern with moderate and high intake but different patterns with light intake, can hardly be explained by confounding effect of any other environmental factor. On the other hand, individuals with a polymorphism in ALDH2 may not drink high amounts of alcohol due to subsequent unpleasant symptoms. Therefore, it is likely that moderate/high intake categories in both regions were mainly comprised of wild type homozygotes with similar alcohol related risk of ESCC.
Only one case-control study reported an inverse association between light alcohol drinking and risk of ESCC in both overall analysis and among never-smokers.54 In that study, the average weekly intake for each type of alcoholic beverages was estimated within each age interval using data collected through self-administered questionnaires. Then the total weekly alcohol intake at each age interval and the total lifetime alcohol intake were calculated. It is therefore possible that participants did not correctly remember their intakes of different alcoholic beverages with various alcohol contents several years in the past. This might have introduced some exposure misclassification, particularly to light and moderate intake levels. When we omitted that study from the light drinking group, the overall association and its heterogeneity was not considerably changed, but the RR among never-smokers moved toward unity (data not shown).
Nevertheless, assessment of light drinking can be difficult even in prospective studies. The reference groups in two prospective studies did not include nondrinkers, because the pattern of association between alcohol intake and risk of ESCC was J-shaped; light drinking was associated with the lowest risk and was set as the reference group.49, 54 As none of the other studies did show a significant inverse association between light drinking and ESCC either in overall analyses or among never-smokers, it is likely that the observed J-shaped association was not valid. This may be related to exposure misclassification and/or confounding effects of other factors in those populations; for example, a part of light drinkers might have considered themselves as nondrinker, or light drinking in those populations might have been an indicator of higher socioeconomic status, which usually shows inverse associations with risk of ESCC.7
Among the cancers that are associated with alcohol drinking,2 the strongest associations have generally been reported with squamous cell carcinomas of the upper aerodigestive tract.7 We found an ∼3.5-fold and 5-fold increase in risk of ESCC associated with high alcohol drinking in prospective studies and in overall analysis, respectively. The overall RRs (95% CI) for oral and pharyngeal cancers among heavy versus non/occasional drinkers have been reported as 4.64 (3.78–5.70) and 6.62 (4.72–9.29), respectively.68 The corresponding value for laryngeal cancer is 2.62 (2.13–3.23).69 This pattern of association suggests that in addition to systemic effects, alcoholic beverages have local carcinogenic effects on the epithelium and/or may cause epithelial lesions, which may facilitate contact of the upper aerodigestive epithelium with other airborne or dietary carcinogens.70
The main advantages of the present meta-analysis are inclusion of a large number of studies, including several prospective studies, investigation of the risk associated with light alcohol drinking, examination of the risk among never-smokers, and the conduct of several subgroup analyses. Other strengths are extensive search of the literature and examination of the retrieved material by at least two coauthors. However, combining studies with various designs and from different populations in any meta-analysis of observational studies can sometimes be misleading. The high heterogeneity among studies reporting on moderate and heavy alcohol drinking was mainly related to the magnitude rather than to the direction of the effect. This may be explained by variation in study design, the method and quality of exposure assessment (e.g., face-to-face interviews versus self-administered questionnaires or current versus average lifetime consumption), and differences in alcohol drinking patterns within strata, particularly in the open-ended high intake category.
In summary, moderate and high alcohol intakes in prospective studies were associated with ∼2-fold and 3.5-fold increased risk of ESCC, respectively. Light alcohol drinking was associated with risk of ESCC mainly in studies in Asian countries. Polymorphism in ALDH2 gene, which is associated with higher serum levels of acetaldehyde, is more prevalent in Asian than in Western populations. Although this suggests that individuals with certain genotypes may be more susceptible to effects of light alcohol drinking, further studies are required to investigate the possible effect modification.