Alcohol consumption and risk of cancer: a Mendelian randomization analysis of four biobanks and consortium data

Alcohol consumption has been linked to cancer risk. Evidence is strongest for seven cancer types: breast, colorectum, oesophagus, liver, mouth, pharynx, and larynx. However, evidence supporting a causal effect from Mendelian randomization is inconsistent.

We perform a comprehensive Mendelian randomization analysis to assess whether genetically-predicted alcohol consumption associates with risk of 20 cancers. Such associations would provide supportive evidence for a causal effect of alcohol consumption on cancer risk. We used 95 genetic variants associated with alcohol consumption at genome-wide significance. Primary analyses were conducted in European ancestry participants from UK Biobank (367,643 individuals), FinnGen (500,348 individuals), All of US (169,312 individuals), and Million Veteran Program (451,206 individuals). We also estimated associations in cancer-specific consortia.

No association was observed between genetically-predicted alcohol consumption and overall cancer (odds ratio (OR) per 1 standard deviation increase in alcohol consumption 0.96, p = 0.45). Among the seven highlighted cancer types, we saw a multiply-corrected significant positive estimate for combined head/neck cancer (OR 1.51, p = 0.001), and nominally significant positive estimates for colorectal (OR 1.21, p = 0.035) and oesophageal (OR 1.42 p = 0.033) cancer. For liver cancer, there was a null estimate overall (OR 1.40, p = 0.10), but a nominally significant positive estimate in Million Veteran Program and when using the ADH1B-rs1229984 variant. For breast cancer, there was a null estimate in biobank data (OR 1.09, p = 0.25) and consortium data (OR 0.98, p = 0.84). Conversely, we observed multiply-corrected significant negative estimates for kidney cancer (OR 0.64, p = 0.0003) and endometrial cancer (OR 0.56, p = 0.0006), and nominally significant negative estimates for non-Hodgkin's lymphoma (OR 0.75, p = 0.010), myeloma (OR 0.61, p = 0.014), and some subtypes of ovarian cancer. There was a nominally significant positive association with cancer mortality (OR 1.44, p = 0.003), although this attenuated on adjustment for smoking heaviness. Limitations include potential invalidity of the genetic variants as instruments, limited power, multiple testing, variable cancer detection rates, and unrepresentativeness of the datasets.

We observed moderate-to-weak evidence supporting causal effects of alcohol consumption on risk of head/neck, oesophageal, and colorectal cancer, inconsistent evidence for liver cancer, and no evidence for breast cancer. Overall, human genetic data do not provide evidence that alcohol consumption is a cause of all cancers and suggest there may even be inverse associations with certain cancer types.