Confounding in nutritional epidemiology; is Mendelian randomization the answer?

Observational studies have reported many beneficial effects of vitamin and mineral supplements on cardiovascular and other diseases, but randomized controlled trials have failed to confirm these. This could be due to the failure of statistical adjustment to eliminate residual and unmeasured confounding by known risk factors. A Mendelian randomization study now suggests a similar problem for the observed beneficial effects of moderate alcohol intake. When 500,000 Chinese people were ranked by self-reported alcohol intake, stroke risk showed a U-shaped curve, risk being lowest at 1-2 units per day. However, when subjects were ranked by genetically conditioned intake, risk was lowest in abstainers and went up linearly with intake. Mendelian randomization may be more suitable for elucidating the effects of diet on health than conventional epidemiology.