A limitation of genetic epidemiological analysis when associations are genuinely J-shaped illustrated using a prospective study of alcohol consumption and vascular disease.

Mendelian randomization, a commonly adopted approach in genetic epidemiology, is of value in distinguishing causal from non-causal associations. There are, however, limitations to the conclusions that can be drawn from this approach when the underlying relationship between an outcome and an exposure is non-linear. In this paper we explore a consequence of these limitations, illustrating these using a prospective epidemiological study of vascular disease in 500 000 Chinese people (the Kadoorie cohort).