Using allele scores to identify confounding by reverse causation: studies of alcohol consumption as an exemplar

BACKGROUND: Mendelian randomization (MR) is a form of instrumental variable analysis used to investigate causality using observational data. Another important, although less frequently applied, use of this technique is to investigate confounding due to reverse causality. METHODS: We used a form of reverse MR and data from UK Biobank in a proof-of-principle study to investigate confounding due to reverse causation. Here we focus on the association between alcohol consumption (exposure) and outcomes including educational attainment, and physical and mental health. First, we examined the observational relationship between alcohol consumption and these outcomes. Allele scores were then derived for educational attainment, and physical and mental health, and the association with alcohol consumption (as the outcome) was explored. Sample sizes ranged from 114 941–336 473 in observational analyses and 142 093–336 818 in genetic analyses. RESULTS: Conventional observational analyses indicated associations between alcohol consumption and a number of outcomes (e.g. neuroticism, body mass index, educational attainment). Analyses using allele scores suggested evidence of reverse causation for several of these relationships (in particular physical health and educational attainment). CONCLUSION: Allele scores allow us to investigate reverse causation in observational studies. Our findings suggest that observed associations implying beneficial effects of alcohol consumption may be due to confounding by reverse causation in many cases.