Background
Hepatocellular carcinoma (HCC) accounts for the great majority of liver cancer diagnoses and deaths and has shown an increasing disease burden worldwide in recent decades [
1‐
3]. Previous studies showed that the risk of liver cancer proportionally increased with the amount of alcohol consumption [
4,
5], suggesting that there is no safe level of alcohol consumption for liver cancer [
6]. For example, in a cohort study of nearly 0.2 million Chinese men, there was a positive dose–response relationship between the amount of alcohol intake and the risk of liver cancer [
7]. However, several studies reported that low-to-moderate alcohol drinking was inversely associated or not significantly associated with the risk of liver cancer (or HCC) [
8‐
10]. For example, a meta-analysis of prospective studies reported a suggestive but not statistically significant decreased risk of liver cancer in individuals with moderate drinking (< 3 drinks per day) compared with abstainers [
10]. Findings from the Liver Cancer Pooling Project showed that light-to-moderate drinking appeared to be inversely associated with HCC risk [
8].
The inconsistencies among observational studies may ascribe to many reasons, including different study designs, small case numbers, incomplete adjustment for confounding factors, reverse causation, and failure to measure lifetime use and patterns of alcohol intake [
11]. Moreover, misclassifying prior heavy drinkers who abstain from drinking currently and nondrinkers who suffer from chronic diseases, such as cardiovascular diseases and diabetes, as abstinence may lead to a spuriously increased risk of liver diseases in this population. Alcoholic beverage type may also affect the association of alcohol intake with HCC [
8]. Additionally, genetic factors that are closely linked to liver diseases may modify the association between alcohol use and liver diseases but were rarely considered in previous studies [
12].
To fill these gaps, in the current study, we dissected the correlation between alcohol consumption and the risk of HCC among 329 thousand UK Biobank participants with the full considerations of potential limitations. In addition, we evaluated whether the associations varied by age, sex, alanine transaminase (ALT) levels, alcohol type and intake frequency, or genetic variants of PNPLA3 and TM6SF2.
Discussion
In this prospective cohort study, we reported a J-shaped correlation between the daily pure alcohol intake level and the risk of HCC, indicating that low-to-moderate drinking may be inversely associated with HCC risk. The J-shaped correlation pattern was detected only in subjects who mainly drank wine but not in those who mainly drank beer and spirits and fortified wine, suggesting that the beneficial effect of low-to-moderate drinking may be largely attributed to wine drinking. We also noted that the association between alcohol consumption and HCC varied by sex, age, ALT level, and genetic variants. The observed J-shaped correlation was more marked in the low-risk populations of HCC.
Alcohol consumption (or ethanol exposure) is one of the well-determined risk factors for HCC, and the underlying mechanisms have been extensively investigated [
21,
22]. Although heavy drinking is well associated with the risk of HCC, whether low-to-moderate drinking inversely associated with HCC risk is still under debate. In this case, cohort studies with large sample sizes, long follow-up times, and particularly detailed information on alcohol exposure are warranted. Our study involved nearly 0.33 million subjects and found a J-shaped association between alcohol use and HCC risk, confirming a previous finding that low-to-moderate drinking is inversely associated with HCC risk [
8]. The Liver Cancer Pooling Project involving more than 1.5 million participants revealed that light-to-moderate consumption of any type of alcohol was associated with a decreased HCC risk [
8]. However, in this study, most participants were assessed for alcohol use over the past year rather than lifetime exposure, which may lead to the misclassification of previous heavy drinkers as abstinent. Additionally, the effect of specific types of alcohol on HCC might be entangled by other alcohol because the predominant alcohol type was not assessed for each participant. The Million Women Study involved approximately 1.3 million women and reported a 41% increased risk of liver cancer in the nondrinkers compared to those drinking ≤ 2 drinks per week [
23]. However, the lifelong nondrinkers were not teased out from those who had stopped drinking. Compared to the two largest cohort studies, our study yielded a similar finding and provided a more detailed dissection of the association between alcohol use and HCC. A recent umbrella review of meta-analyses also showed that low alcohol consumption was associated with a 27% (95% CI 2–46%) decreased risk of liver cancer [
9]. In contrast, some studies reported no association or a proportionally increasing correlation between alcohol use and the risk of liver cancer [
10,
24‐
26]. The between-study inconsistencies may attributed to many reasons, including study design (e.g., case–control study vs. cohort study), sample size and follow-up time, genetic background and HCC incidence of study participants, and more importantly, the recall bias due to documentation and calculation of lifetime alcohol use.
In the present study, we found that the putative protective effect of low-to-moderate drinking on HCC may be largely attributed to wine intake. Indeed, moderate wine drinking has been reported to reduce the risk of chronic diseases including liver fibrosis [
27], cirrhosis [
28‐
30], venous thromboembolism [
31], and esophageal disorders [
32]. In addition, we found that low-to-moderate wine drinking decreases the serum levels of ALT and AST, the leading biomarkers for liver injury in clinical practice, compared with the nondrinkers. Previous studies suggested that polyphenols, the bioactive components in wine including anthocyanin, resveratrol, and gallic acid, may have antioxidant properties, including regulating lipid metabolism and microbiota composition, attenuating ethanol-induced oxidative stress, and anticarcinogenesis [
33‐
35]. Moreover, moderate wine drinkers appear to be at lower risk of becoming heavy drinkers and subjects who preferred wine had healthier diets than those who preferred beer or spirits [
36‐
38], which may further explain the “positive role” of wine drinking.
The association of alcohol use with HCC significantly varied by sex, age, ALT level, and genetic variants. The HCC incidence rates were significantly lower in women, people aged < 60 years, subjects with normal ALT levels, and those carrying non-risk genotypes of PNPLA3 rs738409 and TM6SF2 rs58542926 than in their respective counterparts, although the age of HCC diagnoses was comparable. We observed a J-shaped relationship between alcohol consumption and HCC risk in the HCC low-risk populations. However, when excluding participants mainly drinking wine, alcohol use was associated with HCC risk in a positive dose–response manner in all subpopulations except those with the CC genotype of PNPLA3 rs738409. The results further validate the protective effect of moderate wine drinking on hepatic carcinogenesis. Of note, in the high-risk subgroups of HCC, including men, people aged ≥ 60 years, subjects with abnormal ALT levels, and those carrying the risk alleles of PNPLA3 rs738409 and TM6SF2 rs58542926, we observed a linear relationship between alcohol use and HCC risk, even in those mainly drinking wine. Therefore, drinking alcohol should be completely abstained from in these subpopulations.
Although the current evidence implies that the inverse association between low-to-moderate drinking and HCC may be shaped by bioactive components in wine, the health impact of alcohol (or ethanol) also warrants more investigation. Our nonlinear MR analysis suggested a linear correlation between alcohol intake and HCC, indicating that alcohol per se may have no beneficial effect on the liver. This finding was consistent with prior MR studies on the association of alcohol with cardiovascular diseases and NAFLD [
39‐
41]. However, in our previous study, we found that low-to-moderate drinking regardless of alcohol type was inversely associated with liver fat content [
12], which was also observed in mouse models [
42,
43]. In the current study, the J-shaped correlation between alcohol use and HCC in participants carrying the CC genotype of
PNPLA3 rs738409 was consistent among alcohol types. While the findings seem to indicate a potentially positive effect of alcohol [
44], we should bear in mind that the beneficial effect may only show in a proportion of people.
Our study has some strengths. In contrast to previous studies that were mainly limited to subjects with fatty liver diseases or cirrhosis [
45‐
47], we included relatively healthy people to obviate the bias from alcohol drinking behavior. Second, the effect of alcohol intake was assessed in a continuous form in our study instead of a conventional categorical form as in previous studies [
7,
8,
48], thus preventing the loss of data information and enabling more precise estimates based on the additive Cox regression model. Third, the effect of a specific alcohol beverage type on HCC was evaluated with the exclusion of other alcohols. The confounding effect of alcohol types was therefore largely negated.
Given the strengths, our study also has limitations. First, alcohol use was acquired from a self-report questionnaire, which might be subject to recall bias and underreporting in heavy drinkers. However, there are no ideal biomarkers to measure the lifetime exposure of alcohol. Elaborate questionnaires are an acceptable alternative and have been widely used in epidemiological studies [
39,
49]. Second, the case number of HCC was small in our study, despite the large sample size. The small case number undermined the robustness of modeling estimates, representing by the wide confidence intervals in certain subgroups. In this case, our findings should be interpreted with cautions and need be further validated in cohort study with large case number. However, the association pattern between alcohol use and HCC risk remained consistent in the homogeneous subgroups (e.g., low-risk groups of HCC). Moreover, fitting the volume of alcohol consumption in a continuous form instead of a categorical form in the additive Cox regression model to some extent ameliorates the impact of a small case number [
50]. Our findings therefore give novel clues to the association of alcohol consumption with HCC. Third, the mean age of HCC diagnosis in the UK Biobank participants was 62.3 years, which was younger than the average age of HCC diagnosis in the UK (71.2 years) [
51], indicating that the current follow-up time is not sufficient. Thus, a longer follow-up time is warranted to identify more HCC cases, thereby increasing the statistical power. Fourth, HBV/HCV status was available for only a small proportion of study participants. However, for individuals with HBV/HCV status available, there was no association between these potential covariates and alcohol consumption and no overlap between HCC cases and participants who were positive for HBV/HCV, suggesting that HBV/HCV might have little impact on the association between alcohol use and HCC. Finally, our study participants were of European ancestry, thus limiting the extrapolation of our findings to other populations because of the differences in both alcohol drinking habits and genetic background.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.