Estimating the mediating effect of different biomarkers on the relation of alcohol consumption with the risk of type 2 diabetes

Purpose: Moderate alcohol consumption is associated with a reduced type 2 diabetes risk, but the biomarkers that explain this relation are unknown. The most commonly used method to estimate the proportion explained by a biomarker is the difference method. However, influence of alcohol-biomarker interaction on its results is unclear. G-estimation method is proposed to accurately assess proportion explained, but how this method compares with the difference method is unknown. Methods: In a case-cohort study of 2498 controls and 919 incident diabetes cases, we estimated the proportion explained by different biomarkers on the relation between alcohol consumption and diabetes using the difference method and sequential G-estimation method. Results: Using the difference method, high-density lipoprotein cholesterol explained the relation between alcohol and diabetes by 78% (95% confidence interval [CI], 41-243), whereas high-sensitivity C-reactive protein (-7.5%; -36.4 to 1.8) or blood pressure (-6.9; -26.3 to -0.6) did not explain the relation. Interaction between alcohol and liver enzymes led to bias in proportion explained with different outcomes for different levels of liver enzymes. G-estimation method showed comparable results, but proportions explained were lower. Conclusions: The relation between alcohol consumption and diabetes may be largely explained by increased high-density lipoprotein cholesterol but not by other biomarkers. Ignoring exposure-mediator interactions may result in bias. The difference and G-estimation methods provide similar results.