Alcohol, blood glucose, and type 2 diabetes: Mendelian Randomisation with a focus on sex differences

Importance: Observational evidence often suggests a J-shaped curve between alcohol consumption and risk for type 2 diabetes, with sex-discrepant relationships, but it is unclear whether these findings reflect genuine causal relationships. While genetic-based methods can enhance causal inference, existing studies employing these approaches are limited. Objectives: To evaluate the functional form and strength of the relationship between genetically-predicted alcohol consumption and risk of type 2 diabetes.Design: Non-linear Mendelian randomisation analyses of the UK Biobank cohort study, with a focus on sex-stratified results. Linear Mendelian randomisation and non-linear observational analyses were conducted for comparison. Setting: The UK Biobank is a very large, multicentre, population-based cohort study.Participants: Individuals of European ancestry with complete genetic, alcohol consumption, covariate, and outcome data were eligible for inclusion in Mendelian randomisation and observational analyses. Exposure: Genetically-predicted drinks consumed per week.Main Outcomes and Measure: Incident type 2 diabetes and glycated haemoglobin (HbA1c) levels.Results: For the primary analyses, the sample size was 305,614 (52% female; mean age at baseline: 56.8). Non-linear Mendelian Randomisation suggested a J-shaped relationship between alcohol and risk for incident type 2 diabetes. However, the small protective effect was limited to women, not robust across sensitivity analyses, and held only for overweight and obese women. Observational results produced protective effects for both sexes, even for very large weekly drinking volumes, suggesting residual confounding. For non-linear Mendelian randomisation analyses of HbA1c, alcohol was associated with reduced baseline levels in both sexes – a finding robust to sensitivity analyses. Linear Mendelian randomisation analyses produced no clear results in either direction for T2D or HbA1c outcomes.Conclusions and Relevance: While there was some evidence that low-level alcohol consumption may have causal benefits for average glucose levels, there was no consistent evidence for a protective effect of alcohol use against risk for developing diabetes. Importantly, it was not possible to correct for potential bias from all SES-related confounding/pleiotropy, assortative mating, population stratification, and dynastic effects, necessitating that the current findings be interpreted with caution. Clearer is that observational evidence substantially overestimates the benefits, and underestimates the harms, of alcohol on risk of developing diabetes.