Commentary: Mendelian randomization-inspired causal inference in the absence of genetic data

Studying the long-term causal effects of alcohol drinking is notoriously difficult. Epidemiological studies that use conventional analytical approaches are likely to be confounded and affected by reporting/recall bias and reverse causality, specifically in the form of the sick quitter effect (individuals quitting or never starting to consume alcohol due to underlying ill health).1 Decades of observational data showing J-shaped relationships of alcohol with risk of disease and in particular cardiovascular disease,2 fuelled by confirmation bias, have resulted in alcohol policies such that individuals are recommended to drink in moderation, due to putative cardioprotective effects. Critically, randomized controlled trials (RCTs) to investigate the long-term effects of alcohol drinking are not feasible for reasons including lack of suitable and ethical interventions and extended duration (and hence cost and likely high loss to follow-up).