Low Levels of Alcohol Consumption and Risk of Intestinal Metaplasia: A Cohort Study

Gastric cancer remains the fifth most common cancer and the third leading cause of cancer-related death worldwide despite the decline in incidence during the past decades (1, 2). Gastric cancer is a multifactorial disease and its well-established risk factors include nonmodifiable risk factors (age and sex) and potentially modifiable risk factors such as Helicobacter pylori (H. pylori) infection, smoking, and low physical activity (2).

Alcohol drinking is an established risk factor in various cancers including cancers of the oral cavity, pharynx, larynx, esophagus, liver, colon, rectum, and breast (3). However, the role of alcohol drinking in the risk of gastric cancer and its dose–response relationship is still controversial (4–11). Most previous studies did not incorporate changes in alcohol drinking status over time, even though drinking habits and other confounders can change over time during follow-up and alcohol drinking can be affected as a result of stomach conditions. Gastric cancer follows a sequential carcinogenesis pathway in the order of chronic gastritis, atrophic gastritis (AG), intestinal metaplasia (IM), dysplasia, and adenocarcinoma (12). A few studies have evaluated the effects of alcohol drinking on gastric precancerous lesions but with inconsistent results (13–21). However, those studies were limited by small sample size (16), cross-sectional design (14, 15, 17–20), different definitions of AG based on pepsinogen (15, 21), lack of detailed information on alcohol drinking, and inclusion of former drinkers in the reference group without consideration of sick quitter effect (13, 18–20).

Therefore, we investigated the dose–response association between alcohol drinking and the development of AG and IM in a large cohort of participants who underwent repeated endoscopic examination as a part of a health screening program. We differentiated between former drinkers and lifetime nondrinkers because the former drinkers may have quit alcohol drinking due to previous health problems including stomach conditions. We also accounted for changes in alcohol drinking status and other confounders during follow-up.

The current cohort study was performed in a subsample of the Kangbuk Samsung Health Study, a cohort study of Korean adults who underwent a comprehensive annual or biennial health examination at the clinics of Kangbuk Samsung Hospital Total Healthcare Screening Center in Seoul and Suwon, South Korea as described previously (22). Upper endoscopy, a screening test for stomach cancer, is widely performed every 1 to 2 years as a routine part of the health check-up program in Korea (23). The present cohort study included 302,021 participants who underwent baseline endoscopy from March 1, 2011 (when lifetime alcohol abstinence history was added as part of the health questionnaire) to December 2017 and who underwent at least one follow-up endoscopy prior to December 31, 2019 (Fig. 1). A total of 99,346 participants were excluded for meeting one or more of the exclusion criteria; the final sample included 202,675 participants in the final analysis.

This study was approved by the Institutional Review Board of Kangbuk Samsung Hospital (KBSMC 2020–03–034), which waived the requirement for informed consent due to the use of deidentified data obtained as part of routine health screening exams.

As part of the health exam, information on demographic characteristics, health behaviors, diet, education level, and medical history were collected by standardized, self-administered questionnaires, as described previously (22, 23). Medical conditions diagnosed by a physician, including H. pylori infection, were assessed via the same questionnaire. Smoking status was categorized into never, former, or current smoker. Physical activity levels were assessed via the validated Korean version of the International Physical Activity Questionnaire Short Form and were categorized as inactive, minimally active, and health-enhancing physically active, as described previously (23, 24).

Lifetime abstainers were defined as those who never drank in their lives besides drinking a ritual sip during certain ceremonies. Current alcohol use was assessed as the frequency of alcohol consumption and the amount of alcohol consumed per drinking day. Average alcohol consumption per day was calculated using the frequency and amount of alcohol consumed per drinking day. First, alcohol drinking status based on current alcohol consumption was classified as nondrinking, light drinking, moderate drinking, heavy drinking, or heavier drinking, which correspond to 0, 0.1–<10, 10–<20, 20–<40, or ≥40 g/day, respectively (25). Then, with consideration of lifetime abstinence history, participants were classified as lifetime abstainer, current abstainer, light drinking, moderate drinking, heavy drinking, or heavier drinking. Current abstainers but former drinkers were defined as those with a previous history of drinking alcohol who were nondrinkers at the time of examination. Binge drinking was assessed using the specific question AUDIT-3 for defining binge drinking (“how often do you have six or more drinks on one occasion?” with five responses as never, less than monthly, monthly, weekly, and daily or almost daily), a part of the Alcohol Use Disorders Identification Test (AUDIT), which was developed by World Health Organization (WHO) as a simple method for screening individuals with harmful alcohol consumption (26). In addition, participants were asked about alcohol flushing, a proxy for aldehyde dehydrogenase 2 (ALDH2) deficiency and were categorized into flushers versus nonflushers (27–29).

Anthropometric parameters and sitting blood pressure (BP) were measured by trained nurses. Fasting blood tests included glucose, insulin, lipid profiles, aspartate aminotransferase (AST), alanine aminotransferase (ALT), gamma-glutamyl transferase (GGT), and high-sensitivity C-reactive protein (hsCRP) as described previously (22, 23). Insulin resistance was assessed with the homeostatic model assessment of insulin resistance (HOMA-IR) equation: fasting blood insulin (uU/mL) × fasting blood glucose (mmol/L)/22.5.

Endoscopic examinations were performed by experienced endoscopists using a conventional white light endoscope (GIF H260, Olympus Medical Systems; ref. 23). During endoscopy, any lesions suggestive of gastric pathology were imaged and described according to location, size, and shape. Endoscopists reported the presence of endoscopically suspected atrophy or IM of background gastric mucosa. Endoscopic AG was defined as thinning, whitish mucosal change or visible submucosal vascular patterns, and endoscopic IM as white plaque-like elevations in antrum and corpus (14). H. pylori infection status was evaluated histologically by Giemsa stain only when deemed necessary, such as in peptic ulcer disease, at the discretion of the endoscopist based on the Korean guidelines (30). The history of H. pylori infection was also identified through self-reporting of physician-diagnosed H. pylori infection on the standardized, self-administered questionnaires.

The baseline characteristics of the study participants are presented according to alcohol intake category (lifetime abstainer, current abstainer, 0.1–<10, 10–<20, 20–<40, ≥40 g/day; ref. 25). The primary endpoint was development of separately endoscopic AG and IM. Each participant was followed from his or her baseline exam until either development of the endpoint or to his or her last health exam conducted prior to December 31, 2019, whichever came first. For analysis of the association between alcohol consumption and incident AG, if AG was identified during follow-up, subsequent observations were not incorporated in the analysis. For analysis of the association between alcohol consumption and incident IM, if IM was identified during follow-up, the case was assumed to have developed IM, and subsequent measurements were not incorporated in the primary analysis. In the case that AG developed before IM occurrence, participants were followed until IM development. The incidence rate was calculated as number of incident cases divided by number of person-years of follow-up. Because the development of endpoint would have occurred at an unknown time point between the two visits, the visit at which the endpoint was diagnosed and the prior visit, a parametric proportional hazards model was used to account for this type of interval censoring (31).

The adjusted HR (aHR) and 95% confidence interval (CI) were calculated for incident AG and incident IM according to alcohol intake category. Models were initially adjusted for age and sex and then were further adjusted for center, year of screening exam, smoking status, physical activity, body mass index (BMI), education level, total calorie intake, history of H. pylori infection, history of diabetes, history of hypertension, and history of cardiovascular disease (CVD; model 1). To explore whether inflammation or insulin resistance mediated the association of alcohol consumption with AG and IM, model 2 was further adjusted for hsCRP and HOMA-IR. Furthermore, we conducted time-dependent analyses in which changes in alcohol drinking status and confounders during follow-up were updated as time-varying covariates in the models. The proportional hazards assumption was assessed by examining graphs of estimated log (-log) survival, and no violations of the assumption were found. To determine linear trends of incidence, the number of categories was used as a continuous variable and tested on each model. To further explore the shape of the dose–response relationship of alcohol consumption with the development of IM, restricted cubic splines with knots were used at the 5th, 27.5th, 50th, 72.5th, and 95th percentiles of alcohol consumption. In a sensitivity analysis, participants in the top 25th percentile for serum GGT level, which is often used as a biomarker for alcohol drinking, were excluded to minimize the potential bias by misreporting or bias by disease-related secondary abstinence (15, 32).

Subgroup analyses were conducted by sex (men vs. women) and smoking status (never smokers vs. ever smokers). Among Koreans and other East Asian populations, the ALDH2 variant is a common genotype and might modify the association between alcohol consumption and risk of AG/IM (27, 28, 33–35); thus, subgroup analysis was also performed by alcohol flushing (no vs. yes). Likelihood ratio tests were used to test interactions between alcohol intake and subgroup characteristics.

Statistical analyses were carried out using STATA version 16.0 (StataCorp LP). A P value less than 0.05 was considered statistically significant.

Table 1 shows the baseline characteristics of the 202,675 participants. Lifetime abstainers were more likely to be older and female and to have a medication history of dyslipidemia and history of CVD, but were less likely to have a high education level. Among participants who were current alcohol drinkers, increasing alcohol consumption was positively associated with BP, glucose, liver enzymes, HOMA-IR, salt intake, total energy intake, and unfavorable lipid profiles, while increasing alcohol consumption was inversely associated with high education level. The baseline characteristics are also presented by sex (Supplementary Tables S1 and S2).

During a median follow-up of 4.5 years (interquartile range, 2.7–6.4 years; maximum, 8.8 years), 4,536 participants developed endoscopic IM (incidence rate, 4.8 per 1,000 person-years) and 64,853 participants developed endoscopic AG (incidence rate, 80.5 per 1,000 person-years). For participants who developed AG, the mean follow-up time since enrollment to development of AG was 3.3 years (median, 3.0 years; interquartile range, 1.9–4.6 years); for those who developed IM, the median follow-up time from enrollment to development of IM was 4.2 years (median, 4.0 years; interquartile range, 2.1–6.0 years). Increased alcohol intake categories were positively associated with increased risk of both AG and IM in a dose–response manner (Table 2). After adjustment for confounders (model 1), the multivariable aHRs (95% CIs) for incident IM comparing alcohol intake of <10, 10–19.9, 20–39.9, and ≥40 with lifetime abstainers as a reference were 1.27 (1.03–1.57), 1.35 (1.08–1.68), 1.50 (1.20–1.87), and 1.54 (1.23–1.93), respectively, and the corresponding aHRs (95% CIs) for incident AG were 1.08 (1.03–1.13), 1.15 (1.09–1.20), 1.20 (1.14–1.26), and 1.24 (1.18–1.31), respectively. Current abstainers (or former drinkers) were also at a higher risk for incident AG and IM compared with lifetime abstainers. Adjustment for hsCRP and HOMA-IR qualitatively did not change the results and the association of alcohol intake with AG and IM remained significant (Table 2, model 2). A total of 3,507 (1.7%) participants had information on alcohol consumption only at baseline; the remaining of the participants (98.3%) had at least two data on alcohol consumption at baseline and during follow-up. The participants with alcohol data at only baseline were more likely to be female, former drinker (current abstainer), have diabetes and history of CVD, and were less likely to be highly educated and heavy drinkers (Supplementary Table S3). The average alcohol consumption changed over time (Supplementary Table S4). Most lifetime abstainers (98.6%) maintained a constant state, but a small proportion (1.4%) started drinking. Among former drinkers, about half remained the same state, while the other started drinking again. A majority of light drinkers (82.2% of alcohol category 0.1 to 10 g/day) remained in a constant state, whereas heavy drinkers tended to reduce alcohol drinking over time. When changes in alcohol drinking status and confounders during follow-up were updated as time-varying covariates (Table 2, time-dependent model), the positive associations of alcohol category with incident AG and IM were slightly attenuated but remained significant even in the low-level drinking category defined as 10 to <20 g/day.

In spline regression analyses, there was a dose–response relationship between alcohol consumption levels and the development of AG and IM (Figs. 2 and 3). All drinking patterns including higher frequency of drinking, higher quantity of alcohol consumption per drinking day, and higher frequency of binge drinking were significantly associated with a higher risk of incident AG and IM in a dose-dependent manner (P_trend < 0.05; Supplementary Table S5). The significantly increased risk of incident AG and IM was observed from 1 to 2 times drinking per week and 1 to 2 drinks per day.

We performed analyses among never smokers to address the residual confounding by smoking and found similar results (Table 3); the association between alcohol category and incident AG/IM did not significantly differ by smoking status. The association between alcohol category and risk of IM was similar in men and women without significant interaction (P_interaction by sex = 0.66), whereas the association with risk of AG tended to be stronger in men than in women (P_interaction by sex = 0.03; Supplementary Table S6). Development of IM and AG by alcohol drinking did not statistically differ by the presence or absence of flushing, but the association tended to be stronger among participants with flushing (Supplementary Table S7).

In an additional sensitivity analysis, excluding participants with serum GGT of above its upper quartile (33 IU/L in this study population) did not substantially change the results (Supplementary Table S8).

In this large cohort study with repeated measurements of upper endoscopy, the risk of incident IM and AG increased in a dose–response manner as alcohol consumption increased and this association was consistently observed in time-dependent analyses in which changes in alcohol drinking and other confounders were updated as time-varying covariates. All aspects of alcohol drinking including the drinking quantity, frequency and binge drinking were significantly associated with increased risk of incident IM and AG in a dose–response manner, and this association started from low-level alcohol drinking.

The impact of alcohol drinking on gastric cancer risk remains controversial. Studies addressing this subject have reported inconsistent results, varying from a positive association (4, 8–10) to no evidence of association (11, 36). In 2016, the World Cancer Research Fund suggested that consumption of ≥3 alcoholic drinks (equivalent to ≥45 g/day) is a probable cause of gastric cancer (37). However, the relationship of nonheavy drinking with gastric cancer risk remains still unclear. Studies have reported a positive association of gastric cancer with heavy alcohol drinking (7, 9, 10, 38) and found a lack of association with moderate drinking (4) and even protective effect of light drinking on gastric cancer in women (7). These discrepant findings across the studies may be partly explained by different study populations with different ethanol metabolism capacities and differences in definitions of reference, alcohol measures, and subtypes of gastric cancers (7, 9, 10, 38). Furthermore, during the long duration between alcohol exposure at baseline and gastric cancer diagnosis, individuals with stomach lesions including precancerous lesions may be more likely to abstain from alcohol as a result of their gastric diseases; however, most studies have used a single measure of alcohol consumption at one time point (39). Precancerous lesions can help understanding of the dynamic pathogenesis process and identifying risk factors for precancerous lesions can also help improve our understanding of etiologic risk factors of corresponding cancers (40). Until now, the impact of alcohol drinking on gastric precancerous lesions and its dose–response relationship has remained unclear (13–21). Recently, the American Gastroenterological Association Technical Review demonstrated that the random-effects pooled unadjusted relative risk of having gastric IM in current versus former or never alcohol users was 1.29 (95% CI, 1.12–1.50; ref. 41). However, all studies were limited by uncertain temporal relationships due to cross-sectional study designs; inconsistent differentiation among current, former, and never alcohol users; and lack of adjustments for confounding factors (13, 14, 18, 19, 41). In our detailed dose–response analysis with differentiation from lifetime abstainers versus current abstainers (former drinkers), alcohol drinking at even a light drinking level (<10 g/day) was consistently associated with increased risks of developing both AG and IM. These associations were consistently observed when updated status of alcohol drinking and other confounders over time were treated as time-varying covariates. Smoking is a well-known risk factor for gastric cancer and has been reported as an independent risk factor for gastric IM (23). Even when we performed analyses among never smokers to address the residual confounding by smoking, the association between alcohol drinking and incident precancerous lesions was consistently observed. Furthermore, all measured aspects of drinking patterns including frequency, quantity, and binge drinking were associated with increased risks of incident IM and AG in an independent dose–response manner, and these associations also started from light to moderate drinking.

Although the mechanism for alcohol-induced gastric carcinogenesis has not been fully understood, some suggested mechanisms are as follows: (i) genotoxic effects of ethanol itself or acetaldehyde, the first metabolite of ethanol and an evident carcinogen; (ii) cytochrome p450 2E1 (CYP2E1)-mediated production of reactive oxygen species (ROS) can activate signaling molecules involved in inflammation, angiogenesis, and cell migration as well as cause DNA damage; (iii) folate deficiency can provoke aberrant DNA methylation; (iv) the action of alcohol as a solvent for tobacco carcinogens; and (v) immunosuppression induced by long-term alcohol abuse (42–45). Alcohol consumption could contribute to both the occurrence of alcohol-induced gastritis and the promotion of gastric neoplasm development in the presence of underlying AG (13). Further research is needed to determine how alcohol drinking affects the entire gastric carcinogenesis.

Despite the many strengths of our study, there also are several limitations. First, alcohol drinking was assessed using self-administered structured questionnaires without an objective marker of alcohol consumption, such as phosphatidylethanol or carbohydrate-deficient transferrin, which was similar to most epidemiologic studies (46). Self-reported alcohol intake might be underestimated; however, after excluding participants with serum GGT above the 75th percentile, a commonly used marker of alcohol use, to minimize the potential bias by self-misreporting in drinking (15, 32, 47), the relationship between alcohol drinking and incident AG and IM supports an association of nonheavy drinking with gastric precancerous lesions. Previous studies have addressed the differential effect of types of alcoholic beverages on gastric cancer and reported that wine intake might lower the risk of gastric cancer, unlike that of beer or spirits (38, 48). However, in Korea, a culture of mixing two or more alcohol drinks called “drinking of bomb cocktails” is common, which makes it difficult to investigate the effect of each type of alcohol (49). Although our study focused on average alcohol consumption and drinking patterns relating to quantity, frequency, and binge pattern consumption, beverage type can also differently affect gastric precancerous and cancerous lesions, and we suggest that the effects of these different types of alcoholic beverage require further study. Second, AG and IM were diagnosed based on endoscopic findings but not on histologic findings. However, endoscopically diagnosed AG and IM have been reported to be in acceptable agreement with histologically diagnosed AG and IM and to be associated with gastric cancer development in previous reports (13, 50–54). Even so, varying degrees of sensitivity and specificity of endoscopic AG/IM have been reported: specifically, the sensitivity and specificity of endoscopic AG compared with histologic AG were 65.9% and 58.0% for antrum, 71.3% and 53.7% for corpus, respectively, whereas the sensitivity and specificity of endoscopic IM diagnosis based on histology were 24.0% to 24.2% and 88.0% to 91.9%, respectively (54–56). Third, the intra- and interobserver reliability tests among endoscopists were not performed in this study. Endoscopists involved in a routine health check-up program were unaware of the study purpose and individual's alcohol drinking status while performing procedures; thus, the possibility of misclassification on AG/IM might not have differed by alcohol drinking status. Fourth, H. pylori infection status was defined as self-report of physician-diagnosed H. pylori infection. In only a small proportion (<5%) of participants, H. pylori infection status was evaluated histologically at the discretion of the endoscopists, because a consensus treatment guideline in Korea covered only the treatment of peptic ulcer disease (including scar lesions), early gastric cancer, and marginal zone B-cell lymphoma (MALT type) during the study period (30), thus limiting our ability to account for histologically confirmed or more accurate measures of H. pylori infection in the analysis. Fifth, data on the intragastric location of AG and IM were not included in this study, although most studies in gastric carcinogenesis divided intragastric location into cardiac/noncardiac or proximal/distal. There is still controversy over the frequent location and histologic type of gastric cancer associated with alcohol intake according to race and study design, and thus further study is needed to clarify the alcohol effect on the entire gastric carcinogenesis (9, 57). Thus, we cannot exclude the possibility of some unmeasured or residual confounding factors in the observed findings. Finally, the subjects included in this study were young and middle-aged Koreans in areas where H. pylori is endemic and the ALDH2 variant is common; furthermore, drinking patterns and the prevalence of gene variants encoding the alcohol-metabolizing enzymes differ among racial/ethnic groups (58). Thus, the applicability of our findings to other age groups, populations with a higher prevalence of comorbidities, and other race/ethnicity groups might be limited. Further research using ALDH2 genotyping is needed to examine definitively whether the association between alcohol consumption and precursor lesions of gastric cancer differs by ALDH2 polymorphism.

In conclusion, in this large-scale cohort study of young and middle-aged Korean adults, alcohol consumption from low-level drinking was consistently and independently associated with increased risks of developing AG and IM in a dose–response manner. Our findings indicate that no level of alcohol drinking appears to be safe in relation to premalignant lesions of gastric cancer in this population. However, our findings should be confirmed in other populations and further research is required to determine whether avoiding alcohol drinking can help reduce the development of precancerous lesions of gastric cancer and subsequently reduce gastric cancer risk.

No potential conflicts of interest were disclosed.

K. Kim: Conceptualization, writing–original draft. Y. Chang: Conceptualization, resources, investigation, methodology, project administration, writing–review and editing. J. Ahn: Resources, data curation, investigation. H.-J. Yang: Resources, investigation. S. Ryu: Conceptualization, resources, formal analysis, supervision, investigation, visualization, methodology, writing–review and editing.

This research was supported by grants from Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (2018R1D1A1B07049620) to S. Ryu. The authors thank our staff members of Kangbuk Samsung Health Study for their hard work, dedication, and continuing support.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.