Alcohol causes an increased risk of head and neck but not breast cancer in individuals from the UK Biobank study: A Mendelian randomisation analysis

INTRODUCTION

Harmful effects of excessive alcohol consumption have been suggested for various diseases, including liver cirrhosis, cancer and cardiovascular disease, as well as risk of injuries through accidents ^1,2,3. In the case of the association of alcohol with cancer, both ethanol and the products of its metabolism from the cells are considered to be carcinogenic ⁴. A large body of work exists on the association between alcohol consumption and different types of cancer in model organisms and population samples ^5–7. Recently, a meta-analysis of the cancer and alcohol observational literature produced a list of seven cancers, including cancers of the: oropharynx, larynx, oesophagus, liver, colon, rectum and female breast, that are associated with heavy alcohol consumption ⁸. These results have been widely publicized and currently form part of the official advice provided by the UK Chief Medical Officers’ Low Risk Drinking Guidelines. According to this consumption of alcohol should not exceed 14 units per week for both men and women ⁹.

There are multiple limitations of relying on observational associations to establish the causal role of alcohol on cancer and estimate its true effect on risk, especially when we are interested in either high or low levels of consumption. Alcohol consumption is associated with other cancer risk factors such as poorer diet high in fats and processed meat, obesity, and most importantly, smoking ^10,11. Furthermore, measurement of alcohol consumption, in the vast majority of cases, relies on self-reported drinking habits. This requires the participants to recall how much they drink per day, week, or month period. Studies looking at the accuracy of self-reported alcohol consumption have found that under-reporting is common ¹² and the inaccuracy of the provided information increases with increasing levels of consumption ¹³. Also, the longer the period required to recall, the less accurate the information provided is, with periods longer than a week considered unreliable ¹⁴. Although these limitations can be addressed through the use of randomised control trials, this is not an option in this case due to logistical and ethical problems. In such cases we can make use of the existing genetic variability as a natural randomised control study ¹⁵. Instead of randomly assigning individuals in cases and controls groups, we rely on genetic variation that affects alcohol consumption to identify those that are likely to consume more alcohol and those that are likely to consume less. As long as the chosen genetic variant is not associated with any of the possible confounders of the association between alcohol and cancer, we can use this variant as an unconfounded proxy of alcohol consumption to test and estimate the effect of the associated level of alcohol consumption on risk of cancer. This methodology is described as Mendelian Randomisation (MR) and has been previously used to estimate the causal effect of low to moderate alcohol consumption on cardiovascular risk using a genetic variant from the alcohol dehydrogenase 1B gene (ADH1B) ³. The ADH1B gene codes for the ADH1B enzyme metabolising alcohol into acetaldehyde. A single nucleotide polymorphism (SNP) found in the ADH1B gene, located at rs1229984, changes arginine to histidine in the encoded protein which results in increased levels of acetaldehyde after alcohol consumption and elicits a negative reaction to alcohol¹⁶.

Here using individual level data for 322,193 people from the UK Biobank study, we aimed to establish and estimate the causal effect of alcohol consumption on the risk of cancer types previously suggested as associated with alcohol consumption.

METHODS

RESULTS

After filtering for poor genotyping and non-white British ethnic background and excluding those reporting changes in their drinking behaviour due to their doctor’s advice or due to illness, 322,193 individuals remained in the sample. The mean age of the sample was 56.8 years (Standard Deviation (SD) 8 years), with participants on average slightly overweight with a BMI of 27.4 kg/m² (SD 4.7 kg/m²). The sample included 54% females, and 60% reported current or previous smoking. The average units consumed were 13 units/week (interquartile range 19 units/week). More than a quarter of the sample (28%) reported no alcohol intake per week, 38% were consuming alcohol below the recommended threshold of intake (1-14 units/week), and 34% above this (>14 units/week). The frequency of the minor T allele of the ADH1B rs1229984 SNP was 2.2%. There was no evidence of association between the variant and any of the common risk factors for cancer (ever smoked p-value = 0.197). Those carrying the minor rs1229984 allele were less likely to exceed the recommended level (p-value = 7.92×10⁻¹³²) and more likely to report no average weekly alcohol consumption (p-value = 4.44×10⁻¹⁰⁴). A summary of the overall sample separated per genotype is provided in Table 1.

Observational associations between alcohol and cancer revealed that increasing alcohol consumption was associated with an increasing risk of cancer, including those of the lower, but not upper, digestive system and head and neck cancers. In terms of specific locations, alcohol was associated with neoplasms of the tonsils, larynx, colon and rectum, as well as breast cancer. Alcohol appeared protective for acute MI, used as a control outcome with a known association. Odds ratios, 95% confidence intervals and p-values can be seen in Table 2. When the association between alcohol and cancer was tested in those drinking but within recommended intake limits, no evidence of association between level of alcohol consumption and cancer or acute MI was observed. In those exceeding the recommended level of alcohol, increasing intake was associated with cancers of the tongue, tonsils, larynx, oesophagus, colon, rectum, liver, and pancreas, as well as all broader categories tested (S Table 1). A probable association was present for stomach cancer but did not satisfy our multiple testing adjusted p-value threshold (S Table 1). No evidence of alcohol association with either acute MI or breast cancer were observed per category. No outcome had opposite direction of association per alcohol category. These results are graphically summarised in SFigure1.

The rs1229984 ADH1B variant showed evidence of dominance of the minor allele (p = 0.0236). Using a dominant model, the variant was strongly associated with alcohol intake (F-statistic 692.2, p-value 5.62×10⁻¹⁰⁶) in the sample. Carriers of the minor T allele were consuming on average 2.1 units/week less that the homozygotes of the common allele. The association of the genetic variant with alcohol was also evident for both of those drinking below (F-statistic = 47.4, p-value = 2.63×10⁻¹³) and above (F-statistic = 69.9, p-value = 1.20×10⁻⁶) the maximum recommended 14 units/week, with a difference of 0.35 and 0.87 units/week less respectively. A genetic variant, or allele score, with an F-statistic > 10 for its association with the exposure is considered a strong instrument for use in MR. Testing the association between the variant and the cancer outcomes though, showed no evidence of association between the rs1229984 ADH1B variant and any of the cancer outcomes tested in the full sample, those reporting alcohol consumption of 1 to 14 units/week, or those drinking more than 14 units/week (STable 2). For those reporting no alcohol consumption per week, a probable association suggesting an increase of risk of rectal cancer with the minor allele was observed, but this is well within our expected false positives range (STable 2). In those exceeding the recommended limit of alcohol consumption, we identified a protective effect of the minor allele of the genetic variant reducing alcohol use and acute MI (Odds ratio 0.61, CIs 0.44-0.85, p-value = 2.95×10⁻³) (STable 2).

The instrumental variable regression estimating the causal effect of alcohol consumption on the risk of cancer in all individuals showed no evidence of a causal effect at our multiple testing adjusted p-value threshold, though a probable association was seen with the risk of larynx cancer (probit coefficient 0.044; CIs 0.004-0.085) (Table 2). No evidence of a causal association between alcohol consumption and cancer risk for those reporting that they drink but on average less than 14 units/week was observed. Those exceeding the suggested limit however showed evidence of causal association between alcohol and head and neck cancer (0.053; CIs 0.033-0.073). Probable associations were also seen with an increase of the risk of larynx cancer (0.049; CIs 0.001-0.097), and a decrease of the risk of tongue cancer (−0.47; CIs -0.082 --0.012) with alcohol, though their respective p-values do not reach our multiple testing adjusted threshold (STable 3). For risk of MI we identified evidence of association with alcohol consumption in those drinking more than the recommended maximum (0.048 CIs 0.033-0.062) but not for those drinking below it or in the sample overall (STable 3). The results are visually summarised in Figure 1.

• Download figure
• Open in new tab

Figure 1

Results of 1-Sample MR (ivprobit) analysis for the relationship between alcohol consumption in units/week and cancer in light (≤14 units) and heavy (>14 units) drinkers and in all participants

The two sample MR using the recently published polygenic associations with alcohol consumption ²⁴ did not fundamentally change our results (Table 2). The biggest difference was the probable causal association of any cancer category with alcohol consumption (OR 1.45, CIs 1.12-1.89), though MR Egger results did not support this effect. Alcohol had a probable causal association to head and neck cancer for both the inverse variance weighted (OR 3.86, CIs 1.33-11.21) and MR Egger method. Similarly, cancer of the larynx had again evidence of a probable causal association not reaching our multiple testing adjusted p-value (OR 15.22, CIs 1.65-140.12). Cancer of the oesophagus also showed probable evidence of a causal association (ORs 4.18, CIs 1.02-17.18). Alcohol consumption was not associated with the risk of breast cancer in either of the analyses we performed using events in UK Biobank or when we used outcome associations from the largest breast cancer GWAS available from COGS ²⁶ (OR 0.96, CIs 0.77-1.18) (Table 2 & SFigure 2).

DISCUSSION

We set out to identify and estimate the causal effect of self-reported weekly alcohol consumption on the risk of cancer in the parts of the body in direct contact with alcohol during digestion. We also included breast cancer as a common cancer correlated to alcohol consumption and acute MI as a control outcome for which the causal effect of alcohol was previously well established ^3,27. We found observational associations between weekly alcohol consumption and cancers of the digestive system. The genetic variant chosen as an instrument for alcohol consumption was strongly associated with the exposure but not with the cancer outcomes. In those exceeding the recommended maximum of 14 units/week, alcohol was causally associated with the risk of head and neck cancer. Only limited evidence was available for the causal effect of alcohol to specific cancers and these were focused on cancer of the larynx.

In terms of the observational associations between alcohol consumption and risk of cancer, our result correspond well to what has been previously reported in a meta-analysis of 572 studies ⁸. We confirmed the association of alcohol consumption with an increasing occurrence of cancers in the area of the pharynx (specifically tonsils), as well as colorectal, larynx and breast cancer. We did not replicate the association with oesophageal or liver cancers, except for those consuming more than the recommended maximum of 14 units/week, which also showed an increasing number of cancers of the oral cavity and pharynx, colorectal, pancreas and larynx. Small intestine was not affected by alcohol in any of the categories tested, in both our work and the published meta-analysis. For individuals drinking less than the recommended threshold we did not identify any associations of alcohol consumption with cancer which is in accordance with the published data, except for oral cavity and pharynx ^8,28. A J or U shaped correlation has been suggested between alcohol and risk of cardiovascular disease ^29,30. Here using acute MI we found a protective effect of alcohol in the overall sample that attenuated in magnitude with increasing category of alcohol consumption in accordance with the idea of a non-linear relationship.

The rs1229984 variant we selected as the genetic instrument, based on other published MR work ^3,27,25, was strongly associated with weekly alcohol consumption in our sample. The variant also had other characteristics of a good instrument, such as no evidence of an association of the genetic variant with any of the probable confounders or with any of the cancer types tested in those reporting no alcohol consumption per week. Though, the latter should be considered with caution, since an individual’s report of no weekly alcohol consumption does not rule out the occasional drink. We did not find any associations between the variant and the cancer types tested in the overall sample. The lack of association of the specific variant with stomach cancer agrees with other published data ^22,31,32, though other alcohol associated SNPs have been reported as associated with the outcome. The rs1229984 variant though has been previously associated with aerodigestive cancers, including oral and pharynx, larynx and oesophagus cancers ^33,34,35. Despite not identifying the reported associations, our estimated effects per unit/week were very close to what was reported earlier in a sample of 3,800 cases and 5,200 controls ³³, suggesting that more cancer cases were required to identify some of the associations present. There is some uncertainty for the association of the specific variants with colorectal cancers with some studies reporting an association ³⁶ and others reporting no effect ³⁷. We did not find an association of the variant with any of the lower digestive system cancers. Although the association of the variant used has been previously shown ³ and confirmed in a different population ²⁷ we were only able to replicate the result in those reporting an alcohol intake higher than the recommended threshold.

Although our variant associations failed to identify causal effects reported in the literature, the instrumental variable regression strongly suggested that alcohol consumption higher than 14 units/week was causal to the risk of head and neck cancer and MI. The effect on head and neck cancer appeared to be, at least, partially attributed to an increase risk of cancer of the larynx. These results correspond well to the published associations of the genetic variant, described earlier ^33,34,35, as well as studies showing that laryngeal cancers are reversible following alcohol cessation ³⁸. The observed non-significant increase in the risk of cancer for those consuming alcohol in light and moderate quantities below the recommended threshold has been previously reported ²⁸ though conflicting observational results are also available, especially for breast cancer. We did not identify any evidence of causal association between alcohol consumption and breast cancer, either for the 3,170 UK Biobank or the 122,977 COGS recorded events, irrespective of genetic instrument used. This suggests that the relationship between alcohol and breast cancer ³⁹ is at least severely confounded. Smoking, poor diet and obesity are possible confounders that can also affect the suggested mechanisms for the link between alcohol and breast cancer ⁴⁰.

Limitations exist in our study and they can explain some of the inconsistences we see with published results. The main issue with our data is the lack of statistical power due the relatively small number of recorded events. Although the study population is very large, reaching the 500,000 individuals, it includes people that are more affluent, better educated and in better health than the underlying population of the UK at this age range ⁴¹. This is evident for both the number of cancer events, which are 10-20% lower than expected, and alcohol consumption, with UK Biobank participants less likely to be alcohol abstainers but also less likely to drink alcohol every day compared to the general population ⁴¹. For liver cancer, where alcohol is a well-known major factor for primary hepatocellular cancers ⁴², our results should be interpreted as not covering the directly alcohol associated carcinomas. This is due to our removal from the data of those that changed their alcohol consumption following their doctor’s advice or illness, likely to include these with alcohol related liver disease, and due to that the vast majority of liver cancer cases attributed to cancer cells traveling to the liver from other parts of the body, instead of being primary liver tumours.

To summarise, we showed that alcohol consumption causes cancer of the head and neck and acute myocardial infarction in those exceeding the recommend threshold of 14 units/week. No causal effect of alcohol on cancer was found in those drinking below the threshold. No causal effect of alcohol was observed between alcohol consumption and breast cancer in women, in either the individual level data available here or previously published summary data, suggesting that the observational associations are more likely confounded. Our results contribute to the accumulating evidence for the danger posed from excessive alcohol use for some types of cancer, but not others, and the introduction of the lower 14 units/week threshold as a safe level of alcohol consumption in terms of cancer risk management.